Optimized separable filters

[stephanemagnenat]

Hello, as part of optimizing the Akaze feature detector in Rust CV, I have written SIMD versions of horizontal_filter and vertical_filter using wide. The improvements are from 20 % to 5 times depending on the kernel and the image size. The computed values are similar to what the imageproc version produce.

I assume that people would be interested in this. I'm happy to do a pull request, but given that my other ones have not received answers yet, I wonder whether the maintainers are too busy, in case I'll add them to Rust CV first.

[theotherphil]

A PR would this would be welcome, thanks.