lized GLM converges for all solvers to correct solution.

We work with a simple constructed data set with known solution.
GLM fit on [X] is the same as fit on [X], [y]
                                     [X], [y].
For wide X, [X', X'] is a singular matrix and we check against the minimum norm
solution:
    min ||w||_2 subject to w = argmin deviance(X, y, w)
r