Encoding: Use the different delimeter to pass packed_rawdata and slice_header based...

Encoding: Use the different delimeter to pass packed_rawdata and slice_header based on VAConfigAttribEncPackedHeaders attribute

Currently the packed_slice_header is optional. And it uses the VAEncSliceParameterBuffer
as the delimeter to decide how to insert the packed rawdata/slice_header
for one slice. This is not convenient under some scenario. For example: some
user hope to be more flexible. When the user is responsible for generating the
packed slice_header, it hopes to use the packed slice_header as the delimeter
to determine how to inser the packed rawdata/slice_header for the given slice.
So the VAConfigAttribEncPackedHeaders attriburation of encoding_context is
used to decide which kind of delimeter.
a. When the VAEncPackedSlice is set when calling vaCreateConfig, it will use
the packed slice_header as delimeter. Of course the packed rawdata should be
parsed before the packed slice_header for one given slice. For exmaple:
for the slice 0: the packed rawdata should be parsed before paring the first
packed slice_header. After one packed slice_header is parsed, it will start
to parse the corresponding data for a new slice.

b. When the VAEncPackedSlice is not set when calling vaCreateConfig, it will
use the VAEncSliceParameterBuffer as delimeter.

V1->V2: Return an error instead of only complaining warning message when packed
slice_header is missing for some slice under the VAEncPackedSlice mode. This
is the suggestion from Gwenole and Sreerenj Balachandran.

Signed-off-by: Zhao, Yakui <yakui.zhao@intel.com>
(cherry picked from commit 9d49a6d693aa6c862467a4a879bc86d9cb98dbe5)