Implement Beta Regression as in Ferrari and Cribari-Neto (2004)

[alizma]

Analogous to the Beta regression given in rstanarm here, have the same feature set. This should be an independent class from the existing GLM as otherwise the API will be very complex. Also, this model has considerable differences from the vanilla GLM. Some examples to consider: here and here.

[WardBrian]

Moved existing work to https://github.com/WardBrian/scikit-stan/tree/feature/15-betareg