Close #10096

In the swscale filter plugin, i_visible_width and i_visible_height fields
are now expected to be provided and greater than 0.

This fixes art display problems with the skins2 in recent builds.

Align seek time on second to avoid overflow on seek.

regression from 808e6123 as calculations
still used with/height instead of visible_width/height

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

The initial-segment-number option allows for passing in the number for the first
segment. This is useful if you want to append to an existing live
stream.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Solve the green line bug during thumbnailing.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

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

Enable video filters using the visible dimension of the picture.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

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

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

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

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

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

split speex and speexdsp

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

Use the revision just before libspeexdsp split

Using lowest PTS or DTS is a bad idea, as lower video packets are sent
ahead of pts, and most audio tracks are sending packets near to pts.
So we had audio packets coming late, with a lower pts, and corrupted
our PCR as we take the lowest pts among tracks.
We can't also guess the full tracks rolls, so updating after a full
roll instead of each packet is useless.
Send time is a reliable base time.

Offsets pts & duration

Reads keyframe flag from payload extensions.
Some formats like DVR-MS ships keyframe info into
payload extensions and not as packet flag. Why make things simple ?