[Bug] Undef localized increment values

[guillaumevernieres]

@JohnSteffen-NOAA and I cornered the issue down to the interpolation of the increment from the 0.5 deg grid to the 1/4 deg grid.

@shlyaeva thinks it's probably OOPS PR 3152 that caused the issue.

From Anna:
Here is a tentative fix: https://github.com/JCSDA-internal/oops/compare/develop...bugfix/incoutput. If someone can test it with the problematic case that would be great (I forgot my yubikey unfortunately and can't log in to ursa)

[shlyaeva]

To close the loop on what went wrong. We used to (and in the bugfix branch will again) create an increment on the specified geometry (in our case background geometry), and then call diff between background and analysis on that geometry. Since it was the same geometry as background, no interpolations were involved. After https://github.com/JCSDA-internal/oops/pull/3152 that behavior changed to creating an increment on inner loop geometry, calling diff between background and analysis on that geometry (involves interpolation from background to inner loop resolution), and then interpolating to specified geometry (back to background resolution). So we went from no interpolations to two interpolations. I didn't realize when I reviewed it that a) control increments were by default created on inner loop resolution and b) diff did optional interpolations inside.

[shlyaeva]

I've opened an oops PR for the proposed fix: https://github.com/JCSDA-internal/oops/pull/3188. Let me know if it fixes the issue, and I'll signal that in the oops PR when we're sure.