Make use of the Patsy library

[null-a]

I made a great deal of use of the (very nice) documentation for Patsy while implementing brmp. As a result the two libraries have a lot of functionality in common. (Particularly around interpreting formulae and coding design matrices.) We ought to think about whether we can avoid this duplication of effort eventually.

As far as I know, the main reason we couldn't immediately switch to depending on Patsy is that it doesn't support mixed effects formulas (pydata/patsy#130). The bambi library side steps this by having the user give separate formulae for the fixed and random effects.

It's worth keeping in mind the fact that brms includes a few extensions to the lme4 syntax to support some of its more advanced features. If we choose to do something similar I guess we'd find ourselves extending the formula language beyond what is currently supported by brmp or Patsy.