Package Python

[garrettwrong]

While I still think there is work to do making the CUDA code faster, the core and bindings are probably mature enough to make a proper python package.

[garrettwrong]

#9 mostly packaged python, but there are still several problems, mainly relating to the combination of python and external C/NVCC. This complicates making a python distribution, readthedocs, tox, and ci. Largely the software is pretty solid, perhaps more so than community packaging software I've been dealing with, so this is somewhat annoying.