Split filter_mb_dir() out of filter_mb().

1% overall decoding speed up for cathedral-beta2-400extra-crop-avc.mp4
no speed change for Aladin.mpg
Benchmarks done on Pentium dual 


git-svn-id: file:///var/local/repositories/ffmpeg/trunk@16182 9553f0bf-9b14-0410-a0b8-cfaf0461ba5b