Fix to dumbbell initialization in layer mode

[WenhaoChen89]

This PR fixes the initialization of the dumbbell test to run with the layer mode.

1. The initial layer thickness is changed to follow the isopycnal
2. The initial salinity is set up using the 'fit' method.
3. The sponge is added for the layer mode

[codecov]

Codecov Report

Merging #160 (6909c58) into dev/gfdl (05e705d) will decrease coverage by 0.01%.
The diff coverage is 0.00%.

❗ Current head 6909c58 differs from pull request most recent head 55b9eda. Consider uploading reports for the commit 55b9eda to get more accurate results

@@             Coverage Diff              @@
##           dev/gfdl     #160      +/-   ##
============================================
- Coverage     34.01%   33.99%   -0.02%     
============================================
  Files           259      259              
  Lines         70218    70248      +30     
  Branches      13011    13016       +5     
============================================
  Hits          23884    23884              
- Misses        41835    41865      +30     
  Partials       4499     4499              

Impacted Files	Coverage Δ
src/initialization/MOM_state_initialization.F90	17.23% <ø> (ø)
src/user/dumbbell_initialization.F90	0.00% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 05e705d...55b9eda. Read the comment docs.

[Hallberg-NOAA]

Overall this is a useful contribution that is moving in the right direction, but there is a dimensionally incorrect expression (noted in the line-specific comments) that will need to be corrected before this PR can be accept. As written, this would fail the dimensional consistency testing if Z_RESCALE_POWER is not set equal to H_RESCALE_POWER.

[Hallberg-NOAA]

Some of the issues with this code have now been addressed, but the dimensionally inconsistent expression on line 454 of dumbbell_initialization.F90 still needs to be corrected by multiplying h_in by GV%H_to_Z. I would also request that after this change is made, that this contribution should be tested for dimensional consistency by running it with various settings for H_RESCALE_POWER and Z_RESCALE_POWER; the answers should be identical for any values between about -140 and 140 for these factors. For completeness of testing these two parameters should NOT use the same value.

[Hallberg-NOAA]

All of my concerns with this PR have now been addressed, and I see this as a valuable new contribution.

I am, however, recommending that we handle this PR via a squash merge, as it has a relatively high number of commits (9) for a straightforward new contribution, and several of the commits are partial or full reversions of the earlier commits.

[marshallward]

Gaea regression: https://gitlab.gfdl.noaa.gov/ogrp/MOM6/-/pipelines/16147 ✔️

(NOTE: There were multiple errors from the CI, but appear to be network-related and not to do with this PR.)