avfilter/lut: reduce dereference in the inner loop

For rgb, with a 1080p source, 69 to 74fps on core i5(2 core, 1.8GHz),
and 136 to 160 fps on an core i7(4770R, 3.2Ghz)
Changed the yuv code for consistency, even though the performance
increase is not as obvious as rgb

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>