DiD: allow multiple pre/post intervention observations + correctness fixes

[drbenvincent]

Closes #76

• Main new feature is ability to run DiD with multiple pre and post treatment observations.
• This is covered in tests with test_did_banks_simple and test_did_banks_multi
• DiD was in a pretty bad state to be honest. This is now much more cleaned up and improved. So it should be much more correct now. It can be improved even further (specifically with custom priors) when we have Bambi models available.
• Previously we had a treated predictor. This was equivalent to the group:post_treatment interaction multiplication. But we've done away with this and have replaced treated with post_treatment which is is 1 in all post treatment time points. This allows the model formula to mirror the DiD equations more closely, and makes things more explicit.
• The plotting is not yet perfect, but it improved, and semi-robust to the single and multiple pre/post treatment observation cases.

There are clearly more ways in which DiD and the examples can be improved, but I think this PR represents a very good step forward that's worth merging at this point.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB