get rid of the separate "mask" atlas metadata being used by soca

[travissluka]

I previously added a separate mask to u/v so that the answers would not change during one of the large C++/Fortran refactoring PRs. However, because of this there are two masks for the u/v fields, and the mask metadata is used explicitly for soca specific code.
Remove this by having u/v fields use either the normal cell center mask, or the u/v specific mask. (right now interp uses the cell center mask, and RMS calculation uses the u/v specific mask)

per conversation with @fmahebert

for the u/v fields there is today a interp_source_point_mask metadata pointing to (effectively) the array mask_h, and a mask metadata pointing to the array mask_u/v. If we want to consolidate the generic code to use just mask for the metadata, we need to do one of two things: (a) rename soca's mask metadata to something new (so the metadata name is made available for the generic code) and preserve today's functionality, or (b) change soca's behavior so there is just one mask in use for u/v, which if I understand your prior comment may depend on destaggering the u/v fields.

Ultimately, the correct thing to do is to make sure the U/V fields are destaggered, but until that day, I need to pick a single mask to use for all generic operations.

see discussion as to why this is needed, here #1056