Incrementing robust with different cost function

[pavelkomarov]

Made process term sum of squares, and balanced process and measurement terms, as in Aravkin2013: https://jmlr.org/papers/volume14/aravkin13a/aravkin13a.pdf

There is an equation for how to balance terms in the Huber loss case, but $c_2 \to \infty$ as $\kappa \to 0$, where I believe kappa is equivalent to my huberM. That doesn't make sense to me, because as kappa or M goes to 0, huber becomes the 1 norm, so shouldn't $c_2$ approach $\sqrt{2}$? I'll ask @saravkin on Friday along with possible specialized interior point solver stuff.