Landice mesh_gen: switch conservative remap weight generation from ESMF to mbtempest

[Copilot]

Conservative remapping in landice mesh_gen can hang/segfault with ESMF_RegridWeightGen at high resolution. This change transitions that path to mbtempest (TempestRemap parallel flow) while keeping the existing interpolation workflow intact.

• Core change (landice mesh_gen optional interpolation)

  • In compass/landice/mesh.py::interp_gridded2mali(), replaced conservative weight generation via ESMF_RegridWeightGen with:
    1. mbconvert (SCRIP → H5M),
    2. mbpart (parallel partitioning),
    3. mbtempest --type 5 --weights (conservative weights).
  • Preserved existing downstream remap call (interpolate_to_mpasli_grid) and weight-file contract (gridded_to_MPAS_weights.nc).

• Operational cleanup

  • Removes intermediate .h5m files created during convert/partition to avoid workspace clutter.

• Config/docs alignment for mesh_gen

  • Updated landice mesh_gen config comments and matching User/Developer Guide snippets so nProcs is described as controlling mbtempest (not ESMF_RegridWeightGen).

# old
srun -n ${nProcs} ESMF_RegridWeightGen --method conserve ...

# new flow
mbconvert -B source.scrip.nc source.scrip.h5m
mbpart ${nProcs} -z RCB source.scrip.h5m source.scrip.p${nProcs}.h5m
srun -n ${nProcs} mbtempest --type 5 \
  --load source.scrip.p${nProcs}.h5m \
  --load destination.scrip.p${nProcs}.h5m \
  --file gridded_to_MPAS_weights.nc --weights --gnomonic --boxeps 1e-9

Checklist

• User's Guide has been updated
• Developer's Guide has been updated
• API documentation in the Developer's Guide (api.rst) has any new or modified class, method and/or functions listed
• Documentation has been built locally and changes look as expected
• Document (in a comment titled Testing in this PR) any testing that was used to verify the changes

[trhille]

Testing this on the landice/greenland/mesh_gen case with default .cfg settings on a single Perlmutter cpu node timed out after 4 hours, sitting in the srun -n 128 mbtempest step for most of that time (I can't tell exactly how long). Meanwhile, the existing approach using ESMF_RegridWeightGen in compass/main takes only about 9.5 minutes, so something is very wrong with the implementation in this PR.