Consider how to enable wrapping a fortran model compiled in mixed 32-bit / 64-bit mode

[spencerkclark]

Up to this point when wrapping FV3GFS, we have always compiled the model with 64-bit precision for both the physics and dynamical core. While it can be compiled and run with full 64-bit precision (as is done in #1), for efficiency SHiELD is generally compiled and run in "mixed mode," meaning that the dynamical core uses 32-bit precision values and the physics uses 64-bit values. @lharris4 would like to see this eventually; quoting from NOAA-GFDL/SHiELD_build#27 (comment):

Note that we will probably eventually need to support 32-bit/mixed-mode arithmetic if we wish for greater uptake of the SHiELD wrapper, since we typically run SHiELD with 32-bit dynamics for efficiency purposes.

The Python wrapper currently implicitly assumes that numbers are in 64-bit precision when building the template fortran and Cython modules for getters and setters (see the data types specified in places like here or here). At a minimum this is something that would need to be relaxed / configurable to support wrapping a mixed-mode model. There could be other details I am not considering as well, but this issue is meant to track notes / discussion about this feature.