+Add and use optional unscale args to reproducing_sum

[Hallberg-NOAA]

This PR consists of a series of 4 commits that adds optional unscale arguments to the publicly visible reproducing_sum() interfaces in MOM_coms, and then uses them to refactor the routines in MOM_sum_output and 4 other modules to eliminate about 19 dimensional rescaling factors, and to work more broadly in rescaled variables for dimensional consistency testing.

All answers are bitwise identical, but there are new optional arguments to 2 publicly visible interfaces and a new element in the transparent unit_scale_type. This commits in this PR include:

• 193c83ec9 Use reproducing_sum with unscale in 5 places
• 842bbf4da Work in rescaled units in write_energy
• 4d255baf7 +Add RZL2_to_kg element to unit_scale_type
• 9e7e16523 +Add optional unscale argument to reproducing_sum

[marshallward]

1. Since the unscale * array(..) arguments also implicitly create scaled arrays as copies, you might be able to simplify the reproducible_EFP_sum_* functions with a pointer which conditionally points to these rescaled values for the same amount of work + memory usage.

   Something like

   real, intent(in), target :: array(..)
   real, pointer :: summed_array(..)
   
   if (present(unscale)) then
     allocate(summed_array, mold=array)
     summed_array(..) = unscale * array(..)
   else
     summed_array => array
   endif
   

   then eliminate unscale_ and all of the conditional logic in the solver.

2. field_chksum is being redirected to rotated_field_chksum even though no turns argument is provided. rotated_field_chksum is also redefined as some kind of generic rescaling/rotating wrapper for all debugging operations. I am not quite comfortable with globbing these operations together. It might be more appropriate to create a separate wrapper supporting unscale, or maybe just do it in-place?

   (rotated_field_chksum appears to only be used in one specific place of MOM_restart.F90. Perhaps this function is not even necessary?)

[Hallberg-NOAA]

I have updated this PR to eliminate the changes to the rotated_field_checksum() interface. I plan to revisit this later in a separate PR, probably with a different form. This revised version also avoids using an internal variable name ending in an underscore, and it no longer contains any array-syntax math on multidimensional arrays.

[marshallward]

This looks very good now. The change to reproducing_sum_2d is much better than the one I proposed, and no such change was needed to reproducing_EFP_sum_3d. (My suggestion would have probably made it worse!)

[Hallberg-NOAA]

This PR has been tested at https://gitlab.gfdl.noaa.gov/ogrp/mom6ci/MOM6/-/pipelines/25745.