GetBitContext.size is allways multiplied by 8 -> use size_in_bits to avoid useless *8 in a few inner loops

Originally committed as revision 1486 to svn://svn.ffmpeg.org/ffmpeg/trunk