Switch to matrices for NBCs for non-overlap and Schwarz contacts

[lxmota]

The tractions are returned in an array that has to be indexed in an awkward manner instead of 3 x number-nodes as all other quantities. Change that to be consistent.

[ikalash]

So this is the issue that was messing up the non-overlap coupling? Thanks for finding it! How come it worked for contact for Daria's test cases?

[lxmota]

No, not really. The issue is that I think we should do for non-overlap coupling the same things that we are doing for contact while in the contact phase. I ran the cuboids problem with contact and it works, so there is a difference or differences in what we are doing for coupling but I haven't discovered what those are yet.