boundary blending code tries to blend points outside domain & boundary

[SamuelTrahanNOAA]

Description

The boundary blending code in fv_regional_bc.F90 tries to blend data in regions that are neither inside the domain, nor in the boundary. This problem may originate from the model (fv3atm repo) not the dynamical core (GFDL_atmos_cubed_sphere repo) so I'm making the issue here. In particular, so far as I can tell, the fv_regional_bc.F90 does not allocate the arrays used as input to the blending code, but it does make assumptions about how data is stored in those arrays.

To Reproduce:

1. Fix many bugs in the fv_regional_bc.F90 (PR to be listed here soon)
2. Disable the workaround for the blending bug
3. Compile with DEBUG=ON (preferably with gfortran, which tests this better)
4. Run.
5. Witness aborts in the blending code where the code receives NaN; this is the special value used to initialize data outside the boundary regions.

I have only seen this with the "pt" variable, but I did not test extensively.

Additional context

NOAA-GFDL/GFDL_atmos_cubed_sphere#218

Testing:

Tested with hera.gnu and hera.intel as described in the "To Reproduce" section.