+Improved buoyancy flux diagnostics in non-Boussinesq or layered mode

[Hallberg-NOAA]

This PR consists of a series of 5 commits that make the Kd_heat and Kd_salt diagnostics and the various Bflx_dia_... diagnostics work when MOM6 is run in layered mode and temperature and salinity are used as state variables. When temperature and salinity are not state variables, these diagnostics do not make sense, so they are no longer registered as diagnostics and no longer appear in the available_diags files.

The duplicated code the stratification diagnostics in the diabatic driver was consolidated into the new internal subroutine diagnose_strat_Kd_work(), which is now called with all 3 options in the diabatic driver.

This PR also eliminates the separate non-Boussinesq code to calculate N2_salt and N2_temp in diagnose_Kd_work(). The non-Boussinesq code previously used the Boussinesq reference density (via GV%Z_to_H) for several diagnostics, which is inconsistent with not making the Boussinesq approximation. Ironically, the Boussinesq code did not depend on the Boussinesq reference density when run in non-Boussinesq mode, because it used GV%H_to_RZ instead of GV%Rho0. So this commit both simplifies the code and fixes the diagnostics of the stratification in non-Boussinesq mode.

Finally, there are two sets of changes refactoring KdWork_Diagnostics() for better memory use and to limit the unnecessary exposure of arrays that are not intended to be used outside of this routine. This PR includes a commit that revises the interface to KdWork_Diagnostics() to pass in Kd_salt and Kd_heat rather than storing them in the VBF_CS. These arrays are always available when KdWork_Diagnostics() is called, and storing them just leads to an extra 3-d array copy. Internally within the diabatic driver code, these arrays are also being passed into diagnose_strat_Kd_work(), which calls KdWork_Diagnostics(). The final commit moves Bflx_salt, Bflx_temp, Bflx_salt_dz and Bflx_temp_dz from the VBF_CS into KdWork_Diagnostics() as simple local arrays. These were not exposed outside of KdWork_Diagnostics(), so there was no reason for them to have the more persistent memory that their inclusion in the type entailed. As a part of the refactoring of MOM_diagnose_KdWork.F90, it also adds local variables with the combinations of scaling factors that are repeatedly used for spatially integrating the layer-integrated buoyancy fluxes or as conversion factors for work or buoyancy flux diagnostics, and it does some minor clean-up on the syntax of the register_diag_field() calls in KdWork_init() to follow the MOM6 style guide.

All solutions are bitwise identical, but there are (deliberate) changes in some non-Boussinesq diagnostics and other layer-mode diagnostics that were previously just reporting zeros, while other diagnostics are no longer being offered in cases when they would have been nonsensical.

[knlowman]

Are there any updates on this pending pull request, @Hallberg-NOAA and @breichl ? I see it failed some tests.

[breichl]

Hi @knlowman, I have inspected the code and agree with the proposed changes. I am planning to do some tests in OM5 as soon as possible to confirm that the updated diagnostics look right before I approve the PR, but I anticipate things should be okay. I am unsure why there are failed tests, but I do not think that they are problematic.