Do not allocate ustar and tau_mag together

[Hallberg-NOAA]

Modified MOM_surface_forcing_gfdl.F90 and MOM_surface_forcing.F90 to allocate either ustar or tau_mag, but not both, in the forcing and mech_forcing types, depending on whether the model is in Boussinesq mode. Also added tests to convert_IOB_to_forces in the FMS_cap code and to routines in MOM_surface_forcing in the solo_driver code to ensure that only arrays that are associated are set. All answers are bitwise identical, but checksum statements for the unused arrays are eliminated when DEBUG = True.

[codecov]

Codecov Report

Merging #464 (62aeb04) into dev/gfdl (8d628bd) will decrease coverage by 0.02%.
The diff coverage is 15.38%.

❗ Current head 62aeb04 differs from pull request most recent head 565e5b5. Consider uploading reports for the commit 565e5b5 to get more accurate results

@@             Coverage Diff              @@
##           dev/gfdl     #464      +/-   ##
============================================
- Coverage     37.90%   37.88%   -0.02%     
============================================
  Files           270      270              
  Lines         78100    78125      +25     
  Branches      14436    14456      +20     
============================================
  Hits          29600    29600              
- Misses        43123    43146      +23     
- Partials       5377     5379       +2     

Files	Coverage Δ
...ig_src/drivers/solo_driver/MOM_surface_forcing.F90	23.62% <15.38%> (-0.39%)	⬇️

... and 2 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[marshallward]

There are a surprising number of tau_mag/ustar updates with unnecessary if (associate() blocks inside of them. Although the local value of tau_mag is used to compute ustar, it does potentially impact the performance on platforms with poor branch prediction.

There are also instances of such loops which have the stress magnitude nested inside of the expression for ustar. At the very least, there is some inconsistency here.

Although I am unsure about the frequency of this function call and the overall impact on performance, I would recommend rewriting these loops to avoid such if-blocks within the loops.

[Hallberg-NOAA]

I have revised this PR to move all of the if (associated()) checks outside of the i- and j- do loops, as suggested in the review of the original version by @marshallward.

[marshallward]

This all looks more consistent now.

[marshallward]

Gaea regression: https://gitlab.gfdl.noaa.gov/ogrp/MOM6/-/pipelines/20777 ✔️