#1713 Added the possibility of setting user_block_size manually for gpu::matchTemplate function (and gpu::convolve). Added a buffer param into these functions. Removed using of 2^n block sizes when it's not necessary.