wrong fix for cid #1048729
bs_read in __MAX macro forwards read pointer twice

Opus supports up to 255 channels

Something clearly wrong with that code, as pointed by coverity.
Due to lack of samples using the old audio format 57 bytes
is a guess regarding the previous boundary check.

Stopping and restarting the audio unit or device might take a
considerable amount of time. This leads to flushes, and
the user notice a delay for resuming audio.

Auhal does not have a better API for pause, so lets only play silence
for now. This fixes resume delays with analog mode, and might improve
delays for digital mode (through they are still occuring from
time to time here).

close #9536

div by zero in many places

If fp object was missing...

This is a manual revert of 878169fe, which is not needed anymore.
Additionally, it fixes blayback with devices which are not fast
enough for starting up.

close #9886

damn ! cid 1132614/1132615

ref #9957

Even if should not have any impact during execution.

Seems some encoder reuses previous memory area for junk sections.

close #9864

While current spec doesn't make use of > WORD sizes,
the code was following the spec and reading variable length
up to 3 bytes precision and writing it to a (int *)...
This patch fixes signedness too.

Should fix problems like #2243.
After a seek, the updated position could be the keyframe position
choosen by demux, and then lower than A.

macosx: show 'pause iTunes' option in a more accessible place since it is being asked for a lot (close #9954)

Use deinterlace etc filter output to input as user defined video filters
and make sure visible width/height is defined for that.

No functional change.