Optimization: reduce time taking by MP4_TrackClean

Some user report mp4 demuxer close takes a few seconds on Windows. The sample file has 70 000+ chunks, one sample per chunk. It's not a best practice with mp4 container, but not uncommon. The cpu hot point is DestroyChunk(). It can be fixed by using single memory allocation for p_sample_count_dts/p_sample_delta_dts and p_sample_count_pts/p_sample_offset_pts, and use a few bytes of padding in mp4_chunk_t to avoid memory allocation.