Doesn't get stresses right at OBCs

[kshedstrom]

In going from stresses at h points to stresses on the u/v points, it counts on G%masks2dCu to block exterior values, but that is 1 on open boundaries. We need some OBC code to clean up the mess afterwards.

[kshedstrom]

I should add that this leads to an imperfect restart - but only with MERGED_CONTINUITY = True. Dare I say this is the cause of my MERGED_CONTINUITY blow-ups?