Re-factor of MOM_lateral_mixing_coeffs.F90 (VarMix)

[adcroft]

• VarMix was originally for a crude implementation of a Visbeck et al.
  diffusivity formula. It has become a catchall for resolution functions,
  stored slopes and length-scales. It infact has never contained an actual
  diffusivity. The meaning of the parameter USE_VARIABLE_MIXING was original
  "use Visbeck" but has really meant "allocate and calculate everything in
  this module even if not being used". This has now been cleaned up...

• USE_VARIABLE_MIXING still exists but does nothing.

  • It will be made obsolete shortly.

• Visbeck related fields (VarMix@L2v, ...) are allocated only if
  KhTh_Slope_Cff>0 or KhTr_Slope_Cff>0.

• The Eady growth rates are now allocated and calculated if
  either of Visbeck or MEKE schemes are in use.

• The field VarMix%Rd_dx_h is allocated and calculated if either the
  diagnostic is requested or the field needed "Passivity" or the
  resolution functions.

• The wave speed VarMix%cg1 is calculated only if needed but which
  includes resolution fns, Rd_dx, and FGNV stream fn.

  • Closes FGNV streamfunction for running MOM6 in Neverland configuration #522.

• Added logic in tracer_hordiff() to avoid referencing VarMix%L2v or
  VarMix%SN_v when VarMix%KhTh_Slope_Cff==0.

• Added logic in tracer_hordiff() to avoid referencing VarMix%L2v or
  VarMix%SN_v when VarMix%KhTr_Slope_Cff==0.

• This commit affects available_diags and MOM_parameter_doc for some
  experiments but does not changes answers.

[adcroft]

Passed tests: https://gitlab.gfdl.noaa.gov/ogrp/MOM6/pipelines/2063

[Hallberg-NOAA]

This refactoring is a valuable step toward cleaning up the code. I approve these changes, but I do wonder why we use a "create_group_pass" and then a "do_group pass" on lines 170-171 of MOM_lateral_mixing_coeffs.F90, instead of just doing the single line (and in this case more efficient) call to "pass_var".

[adcroft]

... but I do wonder why we use a "create_group_pass" and then a "do_group pass" ...

Agreed. I didn't add it, just carried it forward. There seems to have been a widespread replacement of single pass_var() with group passes. I prefer pass_var() when there's just the one field.