@JohanMabille I wanted to submit this to get your feedback.

I tested that the outputs are identical to within numerical error of Python so I know the math is correct. I'm not clear on if I should be trying to build expressions to represent the algorithms or do as I'm doing and directly evaluate them. I need to add documentation to these as well. What is the preferred documentation tool?

I Implemented it using a Toeplitz matrix rather than a convolution because it can be multithreaded in the future where direct convolutions cannot be easily. Currently it is likely slower than a single threaded direct convolution but faster than direct evaluation of polynomial fit solving using least squares as would be required in the direct evaluation of the SG filter. Need to benchmark at some point. Likely a different PR.

This PR resolves #1

Thanks for any and all feedback!