Add GrassiaIIGeometric Distribution

[ColtAllen]

This distribution was derived in a research note by Hardie & Fader. It's a simple alternative to the Beta-Geometric mixture distribution, and can be easily extended to support static and time-varying covariates.

I started working on this for a CLV model in pymc-marketing, but internal discussions were in favor of adding this to pymc instead because this distribution has broad application. If pymc-extras is the more appropriate repo for this, let me know.

The research note only provides a PMF and survival function. You can walk through my whiteboarding of the log-likelihood and log-CDF in the below image:

[ColtAllen]

Converted to issue: pymc-devs/pymc-extras#438