GFDL MP fast physics calls through CCPP

[climbfuji]

This PR introduces the necessary changes for the GFDL MP fast physics as described in NCAR/ccpp-framework#95. For further details and dependencies please see there.

[grantfirl]

fv_sat_adj_run standard name questions:

1. zvir is named "moist_gas_constant"; I'm not sure that I like this name. Maybe "virtual_temperature_constant" or something? This is actually in physcons.F90 as "con_fvirt". For what it's worth, I have a standard name of "ratio_of_vapor_to_dry_air_gas_constants_minus_one" in the gmtb_scm_physical_constants.f90 table. Whatever is used, we should strive to be consistent.
2. pkz -- what is "pe" in the standard name? potential energy?

[grantfirl]

I'm not familiar with the code before this, but I'm a bit confused by fv_sat_adj_work versus fv_sat_adj_run. It seems that fv_sat_adj_run is the "driver" for the actual scheme that is "fv_sat_adj_work". I'm guessing that fv_sat_adj_run was necessary to have the inputs/outputs be on the same domain as other schemes (i.e. a set of columns instead of a different array configuration)?

[climbfuji]

@grantfirl yes the issue is that the _work routine (original fv_sat_adj) runs over slabs, i.e. individual z-layers. This is not what CCPP is supposed to do (entire columns) and makes it also difficult wrt parallelization (threading).

[climbfuji]

That sounds reasonable to me, will follow what is in SCM.

[grantfirl]

I'll approve with the caveat that it may be necessary to update the mentioned standard names afterwards.

[climbfuji]

UPDATE: I changed the standard_name from moist_gas_constant to ratio_of_vapor_to_dry_air_gas_constants_minus_one_default_kind and from finite-volume_mean_pe_raised_to_the_power_of_kappa to finite-volume_mean_edge_pressure_raised_to_the_power_of_kappa. I could not use ratio_of_vapor_to_dry_air_gas_constants_minus_one from GFS_typedefs.F90 because of a kind mismatch (default in fv_sat_adj, kind_phys in GFS_typedefs.F90). I will raise an issue to discuss this further, since gfdl_fv_sat_adj is compiled with default real = 8, and in addition the new features in ccpp-framework should prevent mixing types by checking them in the caps at runtime.