Add MPI_APPNUM support for MPMD setups

[maximilian-tech]

Add MPI_APPNUM support for MPMD setups.

In certain (albeit seldom) cases one likes to start nekRS in a MPMD setup (e.g. mpirun -np 2 <nekRS> : -np 2 <other_exe> for e.g. coupling applications.
While the actual coupling will probably happen via User Defined Function (UDF), the communicator passed to UDF_Setup0 is a copy of MPI_COMM_WORLD, which spans both executables (<nekRS> & <other_exe>), which is too large/contains too many processes.

This MR splits MPI_COMM_WORLD based on the color MPI_APPNUM, an attribute of MPI_COMM_WORLD, which returns the index of the program on the command line (e.g. 0 for <nekRS> and 1 for <other_exe>). After that split the communicator passed to UDF_Setup0 is the actual "cut down" nekRS communicator.

Hidden behind a CMake option, as there are probably funky launchers that do not set MPI_APPNUM.

This MR also fixes two bugs:

1. Fail, if MPI cannot provide the requested Thread Level
2. Free commGlobal

[stgeke]

I don’t think this is strictly necessary — we already support multiple sessions. For coupling different codes, a dedicated coupling (driver) application can be used. This approach avoids relying on MPMD, which is not consistently supported across all systems.

What are you trying to do?

[maximilian-tech]

I’m using this to couple NekRS to custom post-processing running on another node. It relies on simple MPI point-to-point communication over MPI_COMM_WORLD in the UDF, so I needed this support for my setup -- no additional driver. I’m aware this is a niche use case, but I decided to upstream it anyway.

If not merged the two bug fixes should be merged though.