When we feed the elementary streams directly to
hardware decoders' ES buffers, the amount of space
for each ES stream is usually limited. When A/V
interleaving is bad, that easily leads to lock-ups
because one of the ES buffers (usually the video
buffer) fills up, while the audio buffer is still
empty or low, which leads to stuttering playback
in the best case, or deadlocks (e.g. no preroll
after seeking) in the worst case.

Detect bad A/V interleaving and stop processing
streams that run ahead, until we've caught up
on the lagging streams, then jump back, and
process the packets from the other stream which
we've skipped.

Needs more work and testing, also should be
made configurable e.g. via a property.