petsc-custom: add internal-boundary ownership patch, mpi test, and docs

[gthyagi]

Problem

Internal-boundary assembly in parallel was rank-dependent for PETSc DMPlex boundary workflows used by UW3 (natural BC / boundary-integral paths). This showed up as branch-split benchmark errors (notably at higher MPI counts) even for identical physics/input.

Root Cause

• Boundary facet ownership was not handled consistently across all boundary assembly paths.
• DMPlexComputeBdResidualSingleByKey used support[0] unconditionally instead of support[key.part].
• For part > 0, points with insufficient support were not filtered/guarded consistently.

What This PR Adds

• New PETSc patch:
  • petsc-custom/patches/plexfem-internal-boundary-ownership-fix.patch
• Patch documentation:
  • petsc-custom/patches/PETSC_MR_DESCRIPTION_INTERNAL_BOUNDARY_OWNERSHIP.md
• PETSc build integration:
  • petsc-custom/build-petsc.sh now prefers the new patch, with legacy fallback to the older patch if the new file is absent.
• Regression test:
  • tests/parallel/test_0765_internal_boundary_integral_mpi.py
• Documentation updates:
  • petsc-custom/README.md
  • CHANGES.md

PETSc Fix Summary

In src/dm/impls/plex/plexfem.c:

• Filter SF leaf (ghost) facets in:
  • DMPlexComputeBdIntegral
  • DMPlexComputeBdResidual_Internal
  • DMPlexComputeBdJacobian_Internal
• In DMPlexComputeBdResidualSingleByKey:
  • filter invalid points for key.part > 0
  • use support[key.part] for closure gather/scatter
  • add support-size guards before part access

Validation

• MPI regression test passes:
  • mpirun -np 4 python -m pytest --with-mpi tests/parallel/test_0765_internal_boundary_integral_mpi.py -q
• Benchmark validation (Ex_Stokes_Kramer_latest.py, case1, natural) is stable across np=1,2,4,8.

Patch Application Guidance

• Use one patch only.
• Preferred: plexfem-internal-boundary-ownership-fix.patch
• Legacy fallback: plexfem-ghost-facet-fix.patch
• Do not apply both together (overlapping hunks / duplicate logic).

[gthyagi]

@lmoresi quick clarification: this PR’s plexfem-internal-boundary-ownership-fix.patch supersedes the earlier plexfem-ghost-facet-fix.patch used in this branch. The newer patch includes the ghost-facet fix plus additional part-consistent boundary assembly updates.

I’m happy to leave consolidation of patch history and documentation to your preferred approach.

I will leave it up to you to create a new PETSc MR or add this to your existing PETSc MR.