decodebin3: Don't output bogus GST_MESSAGE_STREAMS_SELECTED

When `is_selection_done` is called, it checks that all the requested streams are
present in the active stream list ...

... except there could very well be a (about to be removed) stream from the
previous selection present.

Therefore filter the list of streams we add to the message by the streams which
are actually requested.

Fixes issues when switching between different stream types (ex: video-only to
audio-only).

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3457>

Use the correct SSRC(s) when routing a RTCB FB FIR

Previously we tried to route an incoming RTCP FB FIR to the correct ssrc
using the "media source" component of the RTCP FB message. However,
according to RFC5104 (section 4.3.1.2) the "media source" SHALL be set
to 0. Instead the ssrc(s) in use are propagated via the FCI data. Now
a specific GstForceKeyUnit event is sent for every ssrc.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3292>

rtpsource: Track the seqnum for senders

RTP source statistics are tracked for local senders by
treating them as a receiver of their own outbound packets.

Accordingly, track the highest packet seqnum so that the
packets-lost calculation generates a sensible number instead
of always reporting -$number_of_packets_sent

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3454>

adaptivedemux2: Add GStreamer to the deps list

Explicitly dep on GStreamer so as not to accidentally
link to the system version in a git build

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3453>

kmssink: Fix compilation without kernel headers

There was a drm/drm_mode.h included added recently, drm/ is usually
referencing the linux kernel header, but we only requires the libdrm
headers to be installed. On top of this, including drm_mode.h is never
needed as its already included by drm.h.

Fixes #1596

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3452>

onnx: Update to OnnxRT >= 1.13.1 API

- Replace deprecated methods
- Add a check on ORT version we are compatible with.
- Add clarification to the example given.
- Add the url to retrieve the model mentioned in the example.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3388>

h265parse: Add the missing timestamp when splitting a frame.

When splitting a frame, the gst_buffer_copy_region() does not copy
the timestamp correctly for sub frames when the offset is not 0.
We still need those timestamps for each output sub frame.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3448>

h264parse: Add the missing timestamp when splitting a frame.

When splitting a frame, the gst_buffer_copy_region() does not copy
the timestamp correctly for sub frames when the offset is not 0.
We still need those timestamps for each output sub frame.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3448>

dvbsubenc: Write Display Definition Segment if a non-default width/height is used

Otherwise it can't be rendered by dvbsuboverlay or ffmpeg at least.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3446>

inputselector: Add drop-backwards property

When sync-streams=true, drop backwards buffers on pad switch.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3256>

inputselector: Fix waiting on sync-mode=clock

Basically copy over what clocksync does, but taking into account that we
have multiple upstream latencies.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3256>

av1dec: Demote rank to secondary

cerbero does not build this plugin for now, and there's altanative
dav1ddec

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3287>

rtspsrc: Don't replace 404 errors with "no auth protocol found"

When getting a "404 Not Found" response from the DESCRIBE request, the
source produced a "No supported authentication protocol was found" error
instead of passing on the 404, which was confusing.

Only produce this error message when we're handling a response of "401
Unauthorized" without a compatible WWW-Authenticate header.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3414>

vaapipostproc: Fix the negotiation failure of some formats

This patch fixes issue https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1565

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3399>

kmssink: add HDR10 infoframe support

If stream has HDR10 metadata and HDMI device EDID supports it, this patch
will set the DRM properties accordingly.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3303>

kmssink: Add skip-vsync property

The legacy emulation in DRM/KMS drivers badly interact with GStreamer and
may cause the framerate to be halved. With this property, users can disable
vsync (which is handled internally by the emulation) in order to regain the
full framerate.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3303>

adaptivedemux2: Don't leak caps in debug statements

Instead just directly use the stream object (which will report the caps)

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3443>

adaptivedemux2: Don't leak tags

If we got them from GstStream, we should unref them when done

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3443>

adaptivedemux: Use gst_clear_tag_list_where applicable

Clearer and ensures fields are reset

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3443>

oggdemux: Don't leak pending seek event

Make sure any pending seek event is released when going back down to READY.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3443>

rtspsrc: Don't leak sticky events

We have incremented the reference 2 lines above, and
gst_pad_store_sticky_event() does not take a reference, therefore release it

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3443>

parsebin: Don't leak parsepad list on shutdown

Free it as it is down in other cases

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3443>

textrender: Don't pass plaintext as pango markup to Pango

Otherwise e.g. & in the text will cause Pango to complain about invalid
markup and render the text incorrectly.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3445>

textrender: Don't blindly forward all events

Use gst_pad_event_default(), which does the right thing by default.
Especially it does not forward text/x-plain caps downstream.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3445>

event: Add transfer none annotation to gst_event_new_stream_collection()

Update the documentation

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3387>

adaptivedemux2: Fix sticky event storage.

Use the new gst_event_type_to_sticky_ordering() method to retrieve
the order that sticky events should be sent / stored in, instead
of assuming it's the event type value.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3387>

pad: Fix sticky event ordering for instant-rate-change

The event type for instant-rate-change events was poorly chosen,
leading to them being re-sent too late and even after EOS.

Add a mechanism in GstPad for the sticky event order to be
different to the value of the event type to fix that up.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3387>

av1decoder: Clear highest_spatial_layer per sequence header

Clear the value to default zero, indicating that no spatial scalability
layer is used.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3430>

av1decoder: Don't error out by dropped OBU

OBU can be dropped if the current layer is not in selected operation
point

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3430>

av1decoder: Fix wrong spatial layer validation

Highest spatial id and temporal id is independent, and should not drop
temporal enhance layer by the previous condition. Note that
the decision for dropping OBU based on operation point is being
handled in gst_av1_parser_identify_one_obu() already.

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1585
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3430>

av1parser: Don't print warning for expected OBU drop

Dropping an OBU which is not in selected operation point is an
expected condition.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3430>

av1parser: Remove impossible condition

unsigned integer cannot be negative

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3430>

fakesrc: avoid time overflow with datarate

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3433>

webrtcbin: Remove queue after rtpfunnel

The original BUNDLE support commit placed a queue after the
rtpfunnel that combines streams, but I don't see a good reason for
it. It has default settings, so if network output is slow might
accidentally store up to 1 second of pending data, increasing
latency.

Remove it in favour of doing any necessary buffering before
webrtcbin. If it turns out that there is a reason for it to
exist, the limits should probably be configurable and small.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3437>

examples/sendrecv: Remove extra unref of webrtcbin

The code now constructs webrtcbin with a floating ref and then
gives it to the pipeline. The extra unref is one too many.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3436>

rtpbasedepayload: Drop redundant reference timestamp buffer meta in RTP depayloaders

Currently, when rtspsrc property add-reference-timestamp-metadata=true,
a downstream rtph264depay element will attach multiple copies of the
same GstReferenceTimestampMeta to the depayloaded media buffers. This
can have signficant performance impacts further downstream in a pipeline
like the following:

    rtspsrc add-reference-timestamp-metadata=true ! rtph264depay ! h264parse ! ... ! rtph264pay ! ...

For example, if there are 10 packet buffers for a frame of RTP H.264
video, each of those packet buffers will contain the same reference
timestamp meta. The rtph264depay element will then attach all 10
metadata to the depayloaded frame. And then later when we payload the
frame buffer again for proxying, we now have 10 more buffers each with
10 instance of the same metadata. Allocating/deallocating 100+ instances
of metadata @ 30fps for multiple streams has a pretty large performance
impact.

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1578

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3431>

webrtc: Improve GstWebRTCStatsType docstring

Fix a typo of peer-connectiion -> peer-connection

Add a link to the w3c RTCStats type for a description
of what each statistics type is.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3434>

webrtcbin: Fix a typo in debug log

transceiever -> transceiver

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3434>

examples/webrtc: Configure payload types

MR 2398 broke the webrtc sendrecv example
by not configuring the payload types, so both audio and video streams
get sent on payload 96.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3434>

qsv: Promote encoder rank to PRIMARY on Windows

QSV is very well integrated with GstD3D11 infrastructure on Windows,
and this is the recommended H/W encoder element over the MediaFoundation
plugins on Intel GPU system.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3393>

video-frame: Avoid using tile width

The tile width in pixel is not always available. Notably for
8L128 10bit format, the tile is 8x128 bytes, and the pixel
format is fully packed. That means that the tile contains at
least 6 pixels per line, but it also hold some bits of the
pixel from the same line on the previous or next tile.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3424>

video: Add arbitrary tile dimensions support

In current tile representation, only tiles with power of two
width and height in bytes are supported. This limitation
prevents adding more complex tiles formats.

In this patch, we deprecate tile_ws and tile_hs from GstVideoFormatInfo and
replace if with an array of GstVideoTileInfo. Each plane tiles are then
described with their pixels width/height, line stride and total size.
The helper gst_video_format_info_get_tile_sizes() that depends on the
deprecated API is also being removed. This can simply be removed as it wasn't
in any stable release yet.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3424>

gst: serialization of GLibDateTime

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2980>

aggregator: Implement force_live API

Setting force_live lets aggregator behave as if it had at least one of
its sinks connected to a live source, which should let us get rid of the
fake live test source hack that is probably present in dozens of
applications by now.

+ Expose API for subclasses to set and get force_live
+ Expose force-live properties in GstVideoAggregator and GstAudioAggregator
+ Adds a simple test

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3008>

splitmuxsink: Avoid assertion when WAITING_GOP_COLLECT on reference context

I have seen a backtrace out in the wild where this happened. Maybe after
receiving EOS and stream-start on the reference context.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3005>

webrtcbin: Add settings for HTTP proxy

Pass this to libnice which has a simple HTTP 1.0 proxy with basic
authentication only.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2867>

tsdemux: Add pad-name to warning for continuity mismatch

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3141>

decodebin2: Minor debug fix for decodepad

decodedad might have their name changed when exposing, causing a race when
trying to get their name without taking a lock. Just use GST_PTR_POINTER in
debug statements instead

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3428>

imagesequencesrc: Don't leak caps

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3428>

ges-launcher: Don't leak string

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3428>

ges-demux: Don't leak string

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3428>

ges-xml-formatter: Plug some leaks

* Don't leak the mainloop
* Don't leak temporary strings
* Don't leak id when searching in hash table

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3428>

ges-project: Don't leak string

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3428>

ges-structure-parser: Don't leak failed strings

We pass the ownership of current_string to the list of wrong strings, it will be
cleared then.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3428>

ges-launch: Don't leak help string

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3428>

validate:  cleanup the use of GST_VALIDATE_API on Windows

Export or import properly the method from GST_VALIDATE_API
with a proper config.h

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3429>

ci: bump image tags so subprojects get updated

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3432>

subprojects: switch libsoup to a file wrap

With fallback_url.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3432>

subprojects: update glib-networking to 2.74.0 and switch to file wrap

Also has a fallback_url for the tarball.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3432>

subprojects: update glib to 2.74.1

With fallback_url for tarball now too.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3432>

hlsdemux2: Expose EXT-X-PROGRAM-DATE-TIME as tags.

This allows an application to use timestamps associated
with fragments.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1417>

hlsdemux: Expose EXT-X-PROGRAM-DATE-TIME as tags.

This allows an application to use timestamps associated
with fragments.

Patch by: Thomas Bluemel <tbluemel@control4.com>

See: https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/issues/195
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1417>

qtmux: use trun with multiple entries in more cases

The only case where we definitely need to write a new trun is when the
data_offset value does not match the end of the list of entries.
Needing multiple trun atoms is required when interleaving multiple
streams together.

All other cases can be covered by adding more entries to the existing
trun atom.

Fixes playback of fragemented mp4 in ffplay and chrome.

Using e.g. mp4mux fragment-duration=1000 fragment-mode=dash-or-mss
and
mp4mux fragment-duration=1000 fragment-mode=first-moov-then-finalise

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3426>

libs: context: use queried value for attrib

Attribute's value should use returned value from get_attribute for
VAConfigAttribRTFormat, since VAProfileHEVCMain10, in AMD Mesa Gallium,
can process either VA_RT_FORMAT_420 and VA_RT_FORMAT_420_10, which isn't
considered in gstreamer-vaapi design, where encoder's src pads will
expose only 4:2:0 color formats but no 4:2:0 10bit. So, this is a workaround
for this issue while new vah265enc is released.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3397>

vabaseenc: Reduce logging noise if finish_frame fails.

Fixes: #1579
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3423>

gst-docs: Fix broken tables in Cerbero docs

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3425>

amfcodec: Promote encoder rank to PRIMARY

This H/W encoder implementation is the recommended element
on AMD system over the MediaFoundation one.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3394>

urisourcebin: Remove un-needed define

This code hasn't use GValueArray in ages

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Use a non-buffering multiqueue for non-streamable URI

Even though buffering is not required, we need to ensure we are dealing with the
interleave (if any) before pushing the elementary streams further downstream.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Minor doc fix

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

uridecodebin3: Don't leak EOS event

Make sure to unref the EOS event in all code paths
when returning GST_PAD_PROBE_HANDLED

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

playbin3: Move gapless to uridecodebin3

This was the intention from the start, just took me a few years *cough* to
actually implement it properly.

Gapless is handled by re-using as much as possible the same decoders and sinks
if present, and only pre-rolling switching at the sources level (with buffering
if/when needed).

In order to enable "gapless" playback, the "next" uri should be set at any time
between the moment the `about-to-finish` signal is emitted and the moment the
current play item is done. Previously this could only be done with the signal
emission.

This new implementation also allows "Instantaneous URI switching". This allows a
much faster way of switching playback entries while re-using as many elements as
possible. To enable this set `instant-uri` property to TRUE, the default being
FALSE.

API: instant-uri properties
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

playbin3: Remove unused code

This was never used, and if ever it's needed at some point for reference it's
available in gstplaybin.c

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

decodebin3: Clear input collection when linked again

The previous collection no longer applies to this input

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

decodebin3: Improve collection merging

If the collections are the same, don't merge them

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

decodebin3: doc cleanup

Remove old/invalid FIXME and cleanup some logs

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

decodebin3: Remove unused variable

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

multiqueue: Protect reconfiguration with a lock

Add a lock to prevent overlapping of request and release
pads, to close a race where multiqueue might try and
add a slot with an id that hasn't quite finished being
removed yet by another thread.

Fix for https://gitlab.freedesktop.org/bilboed/gstreamer/-/issues/5
and https://gitlab.freedesktop.org/bilboed/gstreamer/-/issues/11

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

multiqueue: Handle gapless input

When dealing with gapless input (i.e. streams with changing group-id in
GST_EVENT_STREAM_START), we need to take into account the elapsed
running-time (if applicable) in order to properly calculate levels and output
time. Without doing this all incoming data from future groups would be
considered as being "late" and would be consumed immediately.

This does **NOT** modify the actual segment and buffer times, and is only used
internally.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

decodebin3: Allow re-using inputs

DecodebinInput (and their backing parsebin or identity) are no longer released
when the corresponding sinkpad is unlinked, but when it's released.

The parsebin element will be resetted:
* If incoming caps are incompatible (was the case before)
* Or when unlinking and it was previously pull-based

This opens the way to use decodebin3 with changing inputs (i.e. gapless)

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

uridecodebin3: use urisourcebin parse-streams property

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Improve buffering handling

Introduce the option to have the streams be parsed with `parsebin` for
compatible sources (i.e. which are eligible for buffering in the same way as
before this commit).

By parsing the inputs directly, this allows more accurate buffering control:
* Instead of relying on potential bitrate information coming from somewhere
* and *without* being linked downstream

If `parse-streams` is activated and the stream is eligible for buffering, then a
`multiqueue` will be used on the output of `parsebin` in order to handle the
buffering.

API: `parse-streams`
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

parsebin: avoid bogus processing

If a parsebin pad is already exposed, don't try to rename it or cause extra
processing which isn't needed

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

decodebin3: Avoid parsebin usage if not needed

If the incoming streams are already parsed, there is no need to add yet-another
parsebin to process it *IF* that stream is compatible with a decoder or the
decodebin3 output caps.

This only applies if all the following conditions are met:
* The incoming stream can *NOT* do pull-based scheduling
* The incoming stream provides a `GstStream` and `GstStreamCollection`
* The caps are compatible with either the decodebin3 output caps or a decoder
  input

If all those conditions are met, a identity element is used instead of a
parsebin element and the same code paths are taken.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

decodebin3: DecodebinInputStream: Hold a ref to srcpad

And move the locking outside of the input stream creation function

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Bring high-watermark level in sync with legacy values

It was always 60%. For some reason this crept in during the initial urisourcebin
refactoring.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

decodebin3: Don't leak collection when releasing inputs

And refactor the function slightly for clarity

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

decodebin3: Protect input freeing with the input lock

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

decodebin3: Refactor parsebin output handling

* Instead of creating temporary `PendingPad` structures, always create a
  DecodebinInputStream for every pad of parsebin
* Remove never used `pending_stream` field from DecodebinInputStream
* When unblocking a given DecodebinInput (i.e. wrapping a parsebin), also make
  sure that other parsebins from the same GstStreamCollection are unblocked
  since they come from the same source

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

decodebin3: Store GstStreamCollection from upstream

If upstream provides stream collection, use/store them as soon as possible

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Move adaptive demuxer handling into ChildSrcPadInfo

It's part of that chain of elements

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Only remove components in PAUSED->READY

With the refactoring, we are guaranteed the components are only created in PAUSED

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Refactor ChildSrcPadInfo and OutputSlot usage

Make an explicit topology/tree of structures:
* ChildSrcPadInfo is created for each source element source pad
* ChildSrcPadInfo contains the chain of optional elements specific to that
  pad (ex: typefind)
* A ChildSrcPadInfo links to one or more OutputSlot, which contain what is
  specific to the output (i.e. optional buffering and ghostpad)
* No longer use GObject {set|get}_data() functions to store those structures and
  instead make them explicit
* Pass those structures around explicitely in each function/callback

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Remove duplicate call

We ensure sources are removed in PAUSED->READY->NULL. No need to call it when
creating the source.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Remove pending pad handling

This was needed to support the legacy handling of changing streams (add new
pads, send EOS and remove old pads).

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Only expose pads once activation has completed

The following problem could happen:
* Thread 1 : urisourcebin gets activated from READY->PAUSED
* Thread 2 : some element causes a pad to be added to urisourcebin , which gets
linked downstream, which decides to activate upstream to pull-based.
  * That requires "activating" the pads from PUSH to NONE, and then from NONE to PULL
* Thread 1 : the base class state change handlers checks if all pads are
activated

The issue is that since going form PUSH to PULL requires going through NONE,
there is a window during which:
* Thread 1 : The pad was set to NONE (before being set to PULL)
* Thread 2 : The base class activates that pad (to PUSH)
* Thread 1 : The attempt to "activate" to PULL fails (silently or not)

This is very racy, so in order to avoid that, we make sure that we only add pads
once the transition from READY->PAUSED in the parent classes is done.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Only allow streams-aware adaptive demuxer

We no longer want to deal with elements that use the old-style of stream
switching and instead expose/remove streams as they appear/disappear

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Simplify initial information collection

When iterating existing pads of a source, directly handle them:
* Raw pads are handled directly
* Pads without caps are connected to a typefind
* Other pads are handled via `handle_new_pad()`

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Use iterator function where applicable

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>

urisourcebin: Document and rename fields

Removes ambiguity when reading the code

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2784>