Optimze nearest neighboor scaling implementation. This is used by the sub pictures core. CPU decrease compared to previous version with my test filter was: 66% for YUVA, 65% for I420 and 71% for RGBA.

Mostly patch by  Sergey Volk

Contrib: Attempt to repair .la pathes. Repeating things, because I don't know how to do it better... Anyway, just an attempt

- Put a limit on the SDP size
- Remove check for access_demux (this is not one, and it would crash later if it were anyway)
- Fix description to not assume this handles UDP only (since it does not anymore)

Add support for I422 streams (un tested).

 * deinterlace.c: compile warning fix.