Begin reorganizing code for packaging

[cthoyt]

Closes #2, but still needs lots of help

• [x] Reorganize code so it's in proper directory structure
• [x] Write setup.cfg, slightly different from conda. They're complementary - of course installing RDKit is a pain and most users won't be able to do it without conda. The python packing side will take care of the non-rdkit dependencies. Documentation is at the bottom of the README.md in the fork
• [ ] Add example on how to use code. I can't do this without help because there is currently no documentation.
  • [ ] It would be good to summarize what the package does, why it does it, and have examples on the readme.
  • [ ] From the last reinvent paper, I assume this is going to allow you to train a model for generating new models. How would a user do this? Normally chemoinformatics people don't include information on how to get or preprocess data in their examples, and that's not so great. Let's do better here
  • [ ] How would a user use a pre-trained model? It looks like there are a few in version control, so an example on that would be necessary as well. There's now an example at https://github.com/cthoyt-forks-and-packages/Reinvent/tree/packaging#reinforcement-learning-using-aurora-model, but it relies on a pickled file, which won't work with the code being reorganized because of how paths to files are stored in pickle

[halx]

Closing as most likely out of sync with current developments.