Package code and make notebook more reusable

[cthoyt]

This PR does several things:

1. After working with @KramerChristian, the underlying Nonadditivity Analysis code has now been packaged and deployed to PyPI. We made some powerful updates that make it much easier to call the code directly from Python, so now the code in this package defers to that rather than calling it via shell script
2. It externalizes many parts of the code in the Jupyter notebook and packages it by adding a setup.cfg and setup.py. The README now reflects how the code should be installed (this automates installing mmpdb and nonadditivity, so the instructions are much more simple). This makes it much easier to focus on the workflow rather than the minutae of the pre-processing, though I think that there could be some more updates to improve this further (e.g., explaining what motivates certain filtering).
3. The code for acquiring and loading the example data is now automated using chembl_downloader, so it's now easy to switch the notebook to any chembl assay as well as automatically upgrade to newer versions of ChEMBL. After a bit of work, I was able to automate reconstructing the same dataframe that was in the GZ file, so it is no longer necessary to store in version control either. Closes #1.
4. Updates the image rendering to use SVG instead of PNG and make higher quality pictures with matplotlib/seaborn. Unfortunately, I'm not familiar enough with PIL to update the circle pictures to be higher quality at the moment