Fix layout regression bug in AdvCore with FV3+ADV

[sdrabenh]

This is 0-diff in the stock configuration. However, these changes fix the layout regression issue when the following lines are enabled in AGCM.rc:

DYCORE: FV3+ADV
AdvCore_Advection: 1

[wmputman]

There are more updates needed. I will make a new commit today

[mathomp4]

@wmputman This PR is throwing a failure in the CI testing in GEOS-ESM/FVdycoreCubed_GridComp#262 (which exercises it). It looks like when you build with FV_PRECISION=R8 an interface is wrong:

/root/project/GEOSfvdycore/src/Components/@FVdycoreCubed_GridComp/@fvdycore/model/fv_grid_utils.F90(3184): error #6633: The type of the actual argument differs from the type of the dummy argument.   [PEM]
        p4 = g_sum(domain, pem, is, ie, js, je, ng, area, 1)
---------------------------^

I think the issue is that pem is declared as:

  real, allocatable:: pem(:,:)

and when you compile as R8, real is now REAL64 and g_sum is now a "fixed" REAL32 routine.

[mathomp4]

@wmputman @sdrabenh I worked with @tclune and I think we have a fix for the FV_PRECISION=R8 case. I'm compiling now and then I'll test GEOS to make sure I didn't break anything.

But one of you might either have to test the FV3+ADV case or teach me how to run that!

[github-actions]

Label error. Requires exactly 0 of: Contingent - DNA, Needs Lead Approval, Contingent -- Do Not Approve. Found: Contingent - DNA, 0 diff