Fix cell-family detection on empty MPI ranks

[gthyagi]

Summary

Determine a distributed mesh's cell family collectively when some MPI ranks own no cells.

DMPlexIsSimplex() is rank-local. On an empty rank it reports False, while populated ranks correctly report the actual family. UW3 then constructed different coordinate finite elements on different ranks. That desynchronised PETSc communication and caused a later labeled HDF5 mesh load to hang in PetscSFSetUp_Basic.

Change

• Read the local simplex result on each rank.
• Contribute None from ranks with no cells.
• Collect the populated-rank results and use one cell family on every rank.
• Use that collective result when selecting triangle, tetrahedron, quadrilateral, or hexahedron element metadata.
• Document the distributed construction invariant.

The serial/populated-rank behavior is unchanged.

Regression

tests/parallel/test_0781_empty_rank_mesh_sequence.py constructs one simplex or tensor cell in 2-D and 3-D on eight ranks, forcing seven empty partitions. It verifies:

• identical mesh.isSimplex values on all ranks;
• identical element-family metadata on all ranks;
• successful construction and integration on a second labeled HDF5-backed mesh, reproducing the sequence that previously hung.

Validation

mpirun -np 8 python -m pytest --with-mpi \
  tests/parallel/test_0781_empty_rank_mesh_sequence.py -v

4 passed per rank in about 23 s

This is split from #689 so the generic distributed-mesh correction can be reviewed and merged independently of Eulerian SUPG transport.

Underworld development team with AI support from Claude Code

[gthyagi]

This independent MPI mesh fix is now a prerequisite for completing the Eulerian SUPG integration stack:

1. merge Fix cell-family detection on empty MPI ranks #691 and Make mesh.cell_size() local and partition independent #692 into development;
2. sync feature/eulerian-supg-transport with updated development;
3. refresh and validate Integrate CitcomS predictor-corrector into unified Eulerian SUPG #689 against that feature branch.

#691 itself remains one focused commit, has no outstanding review comments, and both current CI checks pass. Review is the remaining merge gate.