correct another possible buffer overflow where interlaced would

get stuck to 1 but then width/height would change and interlaced
wouldnt be reset ...

Originally committed as revision 8735 to svn://svn.ffmpeg.org/ffmpeg/trunk