Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

cid 403219
ioctl isn't released

cid #1048853

fix cid #1048883

really unlikely (malloc fails)
cid #1049010 #1049009 #1049008 #1049007 #1049006

We have no mapping for > 8 channels.
cid #1062574

cid #1191891

cid #1191888

Grouping of 2 sync calls into one, use of CATransaction lock / unlock instead of being / commit and introduction of GCD
Signed-off-by: Felix Paul Kühne <fkuehne@videolan.org>

Refs #10722, refs #10723. Also fix a small leak on error.

cid 1191737

- Previous algorithm was unable to guess the first packet
pts because of ogg design (some codecs/opus/vorbis have total
sample at page granule and only granule on last packet. You
then need to compute first packet pts backward using packets
duration).
So we were stuck either assuming a start at pts/pcr zero, or
be unable to handle ogg streaming (non zero start) correctly,
or had to always drop the 0..N-1 packets for first page.
We need libvorbis to correcly compute those durations (code
is far more complex than opus's one).
- Refactorized to discard the tons of cases appended
with each new codec.
- lots of other fixes in many places

Because event will rebuild chapter variable.

because make_path strips last segment