* Bit-exact GaussianBlur reworked to use wide intrinsics

* Added v_mul_hi universal intrinsic

* Removed custom SSE2 branch from bit-exact GaussianBlur

* Removed loop unrolling for gaussianBlur horizontal smoothing