I still don't think this should be here. Note: without this code DV encoder
will accept incorrect pix_fmt's and such, but decoder will work much better
with containers not setting the same attributes. It seems like there will
be a generic mechanism for checking such constraints, but if not I can always
resurrect this check for encoder only.
Originally committed as revision 5114 to svn://svn.ffmpeg.org/ffmpeg/trunk