demux: mp4: optimize mp4_chunk memory allocation

1. Avoid dynamic memory allocation for small samples_per_chunk
(less or equal to 2 sample per chunk). Depends on the test file
and OS, the time taking by MP4_TrackClean can be reduced from a
few seconds to dozens of milliseconds.

2. Merge some allocation to single memory chunk. For mp4 files
with a lot of chunks, the time taking by MP4_TrackClean can be
reduced by half.