Add a Gibbs sampler for the Horseshoe prior and a Polya-gamma normal scale-mixture expansion

[zoj613]

This PR adds functions for creating sampler steps for a normal regression under a Horseshoe prior as described in "A Simple Sampler for the Horseshoe Estimator" (Makalic, Enes & Schmidt, Daniel, 2015).
It also implements the Polya-gamma normal scale mixture expansion described in "Data Augmentation for Non-Gaussian Regression Models Using Variance-Mean Mixtures" (Polson, Nicholas G., and James G. Scott, 2013) and the multivariate normal sampler in "Fast Simulation of Hyperplane-Truncated Multivariate Normal Distributions" (Cong, Yulai, Bo Chen, and Mingyuan Zhou, 2017).

[brandonwillard]

Looks great!

I just added a few comments about general refactoring. If you don't mind rebasing, I can make those changes to this branch.

Otherwise, we need to get the coverage working on this repo so we can make sure that checks out, and we can continue to iterate on MCMC/model-level testing. We could also just merge this after coverage checks out and iterate in follow-ups.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (main@276117d). Click here to learn what that means.
The diff coverage is n/a.

❗ Current head 752e701 differs from pull request most recent head 0f5e947. Consider uploading reports for the commit 0f5e947 to get more accurate results

@@           Coverage Diff            @@
##             main        #1   +/-   ##
========================================
  Coverage        ?   100.00%           
========================================
  Files           ?         2           
  Lines           ?        99           
  Branches        ?         6           
========================================
  Hits            ?        99           
  Misses          ?         0           
  Partials        ?         0           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 276117d...0f5e947. Read the comment docs.

[brandonwillard]

OK, I pushed those refactoring changes. Now, we should try to do one other: separate the Horseshoe and Polya-gamma sampler steps.

In other words, we could have helper functions that generate sampler steps for each independently, then we can provide one that combines the two. This allows people to generate Polya-gamma scale-mixture samplers and normal Horseshoe samplers independently. Plus, it will serve as the basis for us to automate the creation of Horseshoe samplers for a wide variety of known scale mixtures.

Also, we need to set up some tests specifically for the PolyaGammaRV in a test_dists.py. Those tests can be really simple, though, because the implementation is just a pass-through to polyagamma. For that matter, it could be just a single test.

[brandonwillard]

This is looking great. I added some minor comments, but, aside from those, we can merge this after squashing.

[zoj613]

@brandonwillard i've squashed the commits and addressed the comments made.