4.x
Merge pull request #8951 from hrnr:akaze_part2
Remove unnecesary operations and calculations in loop body.
do not use the large "score" buffer; now without non-max suppression OpenCL FAST is pretty efficient
Use sincos instead of sin and cos.