Fast CPU/GPU Earth Mover's distance matrices, test functions, and PyPI/Anaconda automation

[sgbaird]

Alternatively, you could just drop in dist_matrix into the code as you see fit. This also has some stuff built-in for fairly automated PyPI and Anaconda publishing. It would be straightforward for me to incorporate some GitHub action magic with it too. i.e. build with PyPI and Anaconda, run pytest, and make a release automatically via:

git tag -a "#.#.#" -m "<commit message>"
git push origin "#.#.#"

I've been sitting on this content for a while so I figured I should make a PR finally. Feel free to pick and choose what you want to integrate.

It's up on PyPI and Anaconda as chem_wasserstein right now (wanted something up soon for mat_discover, so I have what I need. In other words, it's not urgent. For reference, a 10k x 10k distance matrix takes ~10-20 seconds on my GPU.