Expand ax_helm to allow a coupled compute

[timofeymukha]

So here is an initial proposition for adding an interface to ax, which works on 3 vectors in a coupled manner. After our discussions, this follows the easiest path, which does not directly generalize to a different number of vectors, but perhaps this is OK for no to get the stress formulation in. The ax_helm factory now takes a parameter "formulation", which can be either "simple", meaning the simplified stress tensor, for "full", for the full tensor. Naturally, the existing fluid scheme only uses the simple one.

The idea with this PR is mainly to settling on how to introduce the coupled ax_helm, so that we can start putting in GPU kernels into develop.