Add support for MPI_Comm (mpi08) object in capgen

[peverwhee]

Description

The MPI_Comm DDT must be known to capgen to enable passing around variable of type MPI_Comm.

Summary

In trying to incorporate MPI 2008 into CAM-SIMA, @nusbaume ran into an Unknown DDT type, MPI_Comm error (coming from ~line 540 in metadata_table.py).

A few things I noticed when I was trying to understand the underlying issue:

1. Capgen does not use ccpp_types.F90 at all right now (which I believe is what prebuild uses for this purpose).
2. Capgen does not currently support the way that ccpp_types.F90/.meta grabs the MPI_Comm DDT. It expects that all DDTs for which you have metadata will be defined in the Fortran (rather than just used and made public)

Previously closed issue: #390

Possible Solution

I'm not sure of the best solution, but a couple thoughts:

1. Build MPI_Comm into the framework as a DDT that will always be there (though perhaps we'd need a flag to turn this off in case someone is building without MPI).
2. More generally, update the parser/metadata validation to allow for DDT variables that aren't defined, but just used from an external source and made available/public (as in ccpp_types.F90 currently)

Here's the PR that enabled use of the DDT for prebuild: #523

[climbfuji]

When we switched to mpi_f08 in prebuild, the decision was made to make MPI a hard requirement. There is no need to support building without MPI. It's somewhere in the meeting minutes (last year).

[gold2718]

It looks like #390 should not have been closed. In that issue, a switch to capgen was proposed (--mpi_f08 or something similar). Does that sound like a reasonable approach?

[climbfuji]

It looks like #390 should not have been closed. In that issue, a switch to capgen was proposed (--mpi_f08 or something similar). Does that sound like a reasonable approach?

Is it necessary to still support the old MPI interface? It's 2025 and to me it seems like supporting the 2008 implementation of the MPI library is something we can require?

[dustinswales]

@climbfuji I agree. (Plus we discussed this in great detail last year and everybody was okay with it then).

@peverwhee @gold2718 I came across the same issue as @nusbaume with the CCPP SCM. This was my fix to allow MPI_comm, but maybe this is not the best?

[climbfuji]

Thanks for the link @dustinswales. Given that we all agreed back then, we should require MPI F08, make capgen work with it, and drop support for the ancient interfaces that don't even pass Fortran compatible argument checks.