mesh: fix transfer_meshtags_to_submesh sentinel collision, close test gaps

[jhale]

Replaces the numeric_limits::max() "unmapped" sentinel in
transfer_meshtags_to_submesh with an explicit found-flag vector: a
legitimate tag value equal to that sentinel was being silently dropped
from the transferred result.

Adds regression tests for three previously-untested paths: a
cells/celltypes length mismatch across ranks in build_dual_graph (the
same MPI-ordering defect class as WP1, now verified not to hang),
create_submesh with zero selected entities on some ranks (previously
only traced by hand, never executed), and the sentinel-collision fix
above.

Left cpp/test/mesh/generation.cpp's commented-out per-row adjacency
checks disabled: re-enabling them exposes a real discrepancy between
build_tet's cell-vertex order and the hand-written expected tables,
but Topology construction reorders cell vertices during distribution
and no ordering contract is documented for connectivity(tdim, 0), so
it's unclear whether this is a build_tet bug or an outdated fixture.
Left as a follow-up rather than guessing.

Co-Authored-By: Claude Sonnet 5 noreply@anthropic.com

---

_{Stack created with GitHub Stacks CLI • Give Feedback 💬}