Initialize massless layers to 0 thickness so masked-PE results reproduce non-masked about mom6

[adcroft]

Commit 0f7c079 solves the issue of masked PEs changing the APE calculation.

Leaving ticket open since we still need to check that initializing massless layers to ANGSTROM with ANGSTROM=0 works (should do but hasn't been verified).

[adcroft]

This is just an update. I just tried ANGSTROM=0 with a mix of results:

core dump for MESO_025_63L, benchmark, flow_downslope/layer, global, global_ALE/layer, nonBous_global, resting

Answer changes for DOME, Phillips_2layer, double_gyre, flow_downslope/rho, flow_downslope/z, seamount/layer, seamoount/z, single_column, single_column_z

Fortunately, core dumps are easy to debug. I suspect we are using ANGSTROM where we should be using G%H_neglect, in a few places.

[Hallberg-NOAA]

We have been able to use ANGSTROM=0 for all of the test cases in MOM6-examples for over a year now, although this does require that the various ..._2018_ANSWERS run-time flags be set to .false. (which has been the default since PR #1127 was merged into dev/gfdl in June, 2020). Both the change to ANGSTROM = 0 and changing the value of the various ..._2018_ANSWERS flags will change answers, although the solutions are all qualitatively similar.