slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
per context DCT selection

Originally committed as revision 878 to svn://svn.ffmpeg.org/ffmpeg/trunk