The library delivers us pixel width/height relative to display size
so we must compute it accordingly, otherwise we have conflicting data
in i_sar_num/i_sar_den and i_aspect.

(libmpeg2 does not autodetect on that architecture)

mpeg1 files may have only one sequence header.

It can be backported to 1.0 only after being carefully tested.

It should work as the vout never deletes the pictures used by a decoder.

It should fix lost some picture errors that were critical with dvd menus.
It also seems to fix #2803 but I don't really understand why.

The codec defined will only be used by encoder to force one of the types.

 There will be some regressions, I used automatic sed replacement with manual
checks but there was a lot to change.
 It's only the first pass.

This fixes a case where our libmpeg2 module was not extracting
Closed Captions from the user_data, when this user_data was in the
GOP header, instead of in the picture header. This information is
retained until the next picture (date) is handled.

See sample "bob-dvd-closedcaptioning works in ffmpeg but not in libmpeg2.ts"

Those multi-lines are still to be done (it's going to be fun...)
See also 5af4cc85.

It should close #2195.

Since we require libmpeg > 0.3.2 there is no need for this check.

Though the values are the same, this is cleaner, and more 'future' proof

- This is needed for proper seek/pause/fast forward/...
- The decoder/packetizer do not need to scale packet length anymore as the
decoder thread do it.

This reverts commit f18192c6.
It fixed a few streams while breaking a few others one.

It makes DVD (menu) usable (at least #1784, #1880).

Refs #791

Remove silly out of memory message in an out of memory situation and fixed indentation on bool declarations.

This improves readability a lot and prevent us from reinventing the wheel.

Missing some cases that I could not test.
Also ffmpeg/chroma.c is locked, so I can but leave it broken.
Fix is for the remaining modules is obvious and the issue is easy to detect.