Stabilise the codebase somewhat. For example:

• use Bambi for Bayesian linear/GLM models
• add (or not) a statsmodels model backend
• refine the API