Mixed precision fixes for mass conservation

[FlorianDeconinck]

Mass conservation at f32: GMAO did the work of keeping mass conservation for f32 calculation.

Unfortunately those changes will invalidate 8.1.3 version of the FV3GFS we are using for translate test. We need a new strategy and/or data set.

Another task is to move D_SW fluxes (dpx, x/yflux, cx/y in original Fortran) to always be f64 for the GEOS model. And ask NOAA if they also want this change. Original code is in @fvdycore/model/sw_core.F90:d_sw

Parent: GEOS-ESM/SMT-Nebulae#55

---

• We have a set of data on f32 generated from GEOS 11.5.2
• D_SW mixed precision is ported
• Mass conservation fixes for f32 are ported
• Dynamics dpx fix to mass is implemented

[FlorianDeconinck]

Branch: https://github.com/NOAA-GFDL/PyFV3/tree/fix/geos_f32

[FlorianDeconinck]

Live on https://github.com/NOAA-GFDL/PyFV3/blob/develop__GEOSv_11_4_2

[FlorianDeconinck]

DPX from D_SW implemented but somehow not passing (not sure why)

Dry mass roundoff to be implemented in top level dynamics once the remap is in