Cythonize a pure python file

[isuruf]

cc @mbargull

[conda-forge-linter]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[jakirkham]

How much does this help?

[isuruf]

I tried the command you gave time conda create -n rapids_12_cpython_3 -c rapidsai -c nvidia -c conda-forge -c defaults rapids=0.12 python=3.7 --dry-run. 45s vs 35s

[isuruf]

cc @kalefranz, @jjhelmus, @mingwandroid

[mbargull]

I know I tried that around conda/conda#7676 but it didn't gain much at that point.
I meant to give it another spin after conda/conda#9427 which was exactly meant for this use. I just hadn't gotten to it yet. (I want to do this upstream but I hadn't looked at the CI setup of conda's repository, yet.)

Overall, my opinion on it is that if the Cython version builds fine and passes tests, there should be no downside to using it. I can help in some cases a bit. Definitely not for all cases and not terribly significant in most cases, but it still makes sense.
On the same end, more improvements (also memory-wise) will be gained if one also foregoes the Python interfaces of the SAT solvers and with that the creation of thousands/millions of Python tuples and all. I'll look into that this month. What that will give us is a simple _logic.so that you can drop into conda/common and try out (I know it is that easy on Linux and hope on macOS/Windows it works the same with .dyn/.pyd).