improvements to detc filter: -> use of 8x8 blocks rather than 16x16 to better localize the search for interlacing. this helps detect interlacing in very small motions, e.g. mouths in anime. -> removed some redundant conditions in the logic -> looser condition for detecting lacing and more forgiving of slight mismatches between fields from the two telecine frames to make up for quantization noise in low quality encodes. this code is still mostly experimental but probably better than the old version, so maybe it should be backported to 0.90...?