BW measurement must be per stream as receive window/transfert
rate will differ relatively to chunk size. There's no real
way to get an accurate Max BW with such small files, but it
is still accurate to predict the BW for a download of same
size.

A validation cursor is introduced to qualify bitrates.
Cursor evolves within a window of summed download times.
Adds duration to window if > bitrate, and opposite.
  -PROBE_TIME << cursor << PROBE_TIME
Negative values makes harder to select previously unqualified
bitrates.
Stream quality will rank up quality only if next candidate has
filled its own window.
Stream reselection only occurs when cursor <= 0.
In that way, we can tolerate temporary bitrate lowering or increase
and avoid the hiccup reselection problems we had until now.

Gets rid of useless indirect access to chunks.
Uses time as chunk based instead of indexes.
Does not keep chunks on live stream.

Fixes #12722

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Felix Paul Kühne <fkuehne@videolan.org>

(cid #1251048 and #1251059)

Fixes searching for native qt5 progs on cygwin and msys2

The reason for the issue is that on windows if there is a space in the path
or if the program is prepends with .exe and we specify the search PATH
in AC_PATH_PROGS we must enclose them in quotes or the program will not be
found.

For future reference, if anyone uses AC_PATH_PROGS please enclose defined
search paths (if you have any) in quotes
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

This mostly reverts commit 6a6c23bf.

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

They are superseded by makeCurrent() and releaseCurrent().

The documentation promises that the picture is unlocked before it is
displayed, so actually do that.

Cc: Antti Ajanki <antti.ajanki@iki.fi>

date is always >= VLC_TS_INVALID

Original channels is under-specified in my opinion. If you know where
the design documentation is, please tell me.