Fix IndexError in _pic_swarm_in_mesh on MPI ranks with empty cell partition

[hardik-corintis]

Description

Fixes #5077.

In _pic_swarm_in_mesh, parent_cell_nums_local contains -1 sentinels for points not
owned by the current rank. On ranks with an empty cell partition, cell_closure has shape
(0, k) and NumPy rejects -1 as an axis-0 index, causing an IndexError when
interpolating from a Submesh onto its parent mesh in parallel.

Fix: index cell_closure only on visible_idxs (the non-sentinel entries), pre-filling
the rest with -1. The same issue exists in the extruded branch: _parent_extrusion_numbering
floor-divides parent_cell_nums_local, and -1 // N == -1 in Python, so sentinels pass
through into base_parent_cell_nums and cause the same crash. Both branches are fixed.

Two regression tests added in tests/firedrake/submesh/test_submesh_interpolate.py using
8 MPI ranks:

• test_submesh_interpolate_3Dcell_2Dfacet_empty_rank_8_processes: interpolates from a
  Submesh of the x=1 face of UnitCubeMesh(2, 2, 2) onto the parent mesh.
• test_submesh_interpolate_3Dcell_extruded_empty_rank_8_processes: interpolates from
  ExtrudedMesh(UnitSquareMesh(1, 1), layers=3) (6 cells) onto a UnitCubeMesh(2, 2, 2),
  guaranteeing at least two ranks own zero extruded cells.

[leo-collins]

I think this is good. The same thing probably happens for the extruded case too - can you add a test for this (and a fix if necessary)?

[hardik-corintis]

Thanks @leo-collins for the quick review. I have applied all the inline suggestions and added the test and fix for the extruded meshes.

[connorjward]

Should this be going into release?

[hardik-corintis]

Should this be going into release?

If this is a question for me, I am not sure, I would go with your recommendation. My understanding from the contribution guide is that bug fixes generally go into release, but I am happy to follow whatever is preferred here.

[connorjward]

Unless this is somehow relying on features that only exist in main then I think that this should go into release. To set that up I'd recommend rebasing, force pushing and then changing the target branch of this PR. Just changing the target branch of the PR leads to conflicts.

[hardik-corintis]

Done — rebased onto release and retargeted the PR!

[connorjward]

Thanks!