Set wrf.exe namelist values and mpirun arguments from the environment

[aethr]

Stacked on #88, so that the image for this PR carries the [om-metrics] instrumentation as well. It's for the experiments in #89.

Changes

• WRF_NAMELIST_OVERRIDES: Fortran namelist text, applied to namelist.input after real.exe has run. For example:

  &time_control history_begin_h = 12, use_netcdf_classic = .true., debug_level = 0 / &namelist_quilt nio_tasks_per_group = 1 /
  

  It's applied after real.exe so the values reach wrf.exe only. real.exe reads the same file on a single rank and can't set aside I/O servers.
• Quilting: I/O servers come out of the mpirun total, so the default rank count is now safe_mpi_tasks() over the CPUs left, plus the I/O tasks. Setting NCPUS overrides this as before, and must include the I/O tasks.
• MPI_EXTRA_ARGS: passed to wrf.exe's mpirun, for options such as -bind-to core.

With neither variable set, the generated namelist and run.sh behave exactly as before.

Testing

• Unit tests for apply_namelist_overrides() and quilt_tasks(). make test: 42 passed.
• Rendered run.sh from the template and ran it against a stub mpirun. With no variables set it gives -np 64 ./wrf.exe; with MPI_EXTRA_ARGS="-bind-to core" NCPUS=32 it gives -bind-to core -np 32 ./wrf.exe.
• Not yet run end to end. The first wrf.exe spends over half its run writing history output; cut that before sizing ranks #89 scenario in sandbox will be the real test.

🤖 Generated with Claude Code

[aethr]

Closing, as the perf testing this was made for is concluded. Results are in #89 .