Yes, yet another evil all-in-one commit of intervowen changes. I suck.

Yes, yet another evil all-in-one commit of intervowen changes. I suck.
* Drop "state" directory, fold that into "runtime directory"
* No longer automatically rewind when a new stream connects
* Rework sound file stream, to cause a rewind on initialisation, shorten _pop() code a bit
* Fix reference counting of pa_socket_server in the protocol implementations
* Rework daemon initialization code to be compatible with non-SUID-root setups where RLIMIT_RTPRIO is non-zero
* Print warning if RT/HP is enabled in the config, but due to missing caps, rlimits, policy we cannot enable it.
* Fix potential memory leak in pa_open_config_file()
* Add pa_find_config_file() which works much like pa_open_config_file() but doesn't actually open the config file in question. Just searches for it.
* Add portable pa_is_path_absolute()
* Add pa_close_all() and use it on daemon startup to close leaking file descriptors (inspired from what I did for libdaemon)
* Add pa_unblock_sigs() and use it on daemon startup to unblock all signals (inspired from libdaemon, too)
* Add pa_reset_sigs() and use it on daemon startup to reset all signal handlers (inspired from libdaemon as well)
* Implement pa_set_env()
* Define RLIMIT_RTTIME and friends if not defined by glibc
* Add pa_strempty()
* rename state testing macros to include _IS_, to make clearer that they are no states, but testing macros
* Implement pa_source_output_set_requested_latency_within_thread() to be able to forward latency info to sources from within the IO thread
* Similar for sink inputs
* generelize since_underrun counter in sink inputs to "playing_for" and "underrun_for". Use only this for ignore potential rewind requests over underruns
* Add new native protocol message PLAYBACK_STREAM_MESSAGE_STARTED for notification about the end of an underrun
* Port native protocol to use underrun_for/playing_for which is maintained by the sink input anyway
* Pass underrun_for/playing_for in timing info to client
* Drop pa_sink_skip() since it breaks underrun detection code
* Move PID file and unix sockets to the runtime dir (i.e. ~/.pulse). This fixes a potention DoS attack from other users stealing dirs in /tmp from us so that we cannot take them anymore)
* Allow setting of more resource limits from the config file. Set RTTIME by default
* Streamline daemon startup code
* Rework algorithm to find default configuration files
* If run in system mode use "system.pa" instead of "default.pa" as default script file
* Change ladspa sink to use pa_clamp_samples() for clamping samples
* Teach module-null-sink how to deal with rewinding
* Try to support ALSA devices with no implicit channel map. Synthesize one by padding with PA_CHANNEL_POSITION_AUX channels. This is not tested since I lack hardware with these problems.
* Make use of time smoother in the client libraries.
* Add new pa_stream_is_corked() and pa_stream_set_started_callback() functions to public API
* Since our native socket moved, add some code for finding sockets created by old versions of PA. This should ease upgrades

git-svn-id: file:///home/lennart/svn/public/pulseaudio/branches/glitch-free@2329 fefdeb5f-60dc-0310-8127-8f9354f1896f