Robustness Tests

[cicdw]

We should have a testing framework / experimentation engine to test the algorithms for robustness to various types of input data (normalized / not-normalized / bad condition numbers / partial separability / low class sizes, etc.) -- eventually we might even want some sort of switching method that, for example, might detect bad conditioning in Newton's method and switch to gradient descent. This will require first writing a way to generate such data.

@mcg1969 @jcrist

[mcg1969]

Once I get a bunch of delegatable tasks built out for my larger project I'm going to take a stab at this. Hoping to do so Wednesday or Thursday.