Fix two of my stupid bugs that prevented using the queue with transport streams:

Fix two of my stupid bugs that prevented using the queue with transport streams:
 - keep a cache of the pid/substream id mappings from each scan rather than assuming an encode will immediately follow a scan (there will be lots of scans followed by lots of encodes when the queue is used).
 - rewrite a few things to get rid of static variables. hb_ts_stream_decode is called by both scan & reader and they're in different threads. All the working storage & state has to either be in the stream struct or on the stack so it's private to the calling thread or it gets corrupted when doing a scan in the middle of an encode.

git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@1351 b64f7644-9d1e-0410-96f1-a4d463321fa5