RTP packets were previously take from a buffer pool. This was messing
up the Client, since that buffer pool was used by multiple routines at
once, and was probably messing up the Server too, since packets can be
pushed to different queues and there's no guarantee that these queues
have an overall size less than ReadBufferCount.
This buffer pool is removed; this decreases performance but avoids bugs.
* split tracks from medias
* move tracks into dedicated package
* move media into dedicated package
* edit Medias.Marshal() in order to return SDP
* add medias.Find() and simplify examples
* improve coverage
* fix rebase errors
* replace TrackIDs with MediaIDs
* implement media-specific and track-specific callbacks for reading RTCP and RTP packets
* rename publish into record, read into play
* add v2 tag
* rename tracks into formats
