Smoothed Hinge Loss analysis and refactor

[ashelkovnykov]

Smoothed Hinge Loss is a bit of a black sheep currently. It was introduced as an experiment with the goal of approximating an SVM using a GLM with a modified Hinge Loss function. However, it's currently uncertain how well the function converges. In addition, it's treated a bit differently in the code than the other functions (ex. normalization is ignored for it). There are no current use cases for it.

We need to take a closer look at Smoothed Hinge Loss and its future within Photon.