fixed alignment (static variables where sometimes not 8-byte aligned)

added half uv interpolation support
added prefetch
BGR15 support in MMX (untested) (so BGR15,16,24,32 are supported)
special unscaled height version (not much faster but it doesnt interpolate uv vertically)

Originally committed as revision 2317 to svn://svn.mplayerhq.hu/mplayer/trunk/postproc