Implement TASK-030 (Periodic Boundary), Stage 4's ninth and last task

[AdamClemens]

Summary

• Adds StructuredCartesianMesh.wrapped_neighbour_cell (mesh geometry, not a third BoundaryCondition shape, per this task's own recorded Design decision) and threads a new periodic_pairs mapping into FirstOrderUpwindAdvection/CentralDifferenceDiffusion alongside boundary_conditions — no interface change, no new ADR (mirrors diffusion_coefficient's own TASK-024 precedent).
• The round-trip invariant (Stage 4's own Intent for this task) is checked as convergence under mesh refinement, not exact equality at one resolution — a genuine numerical finding, verified before writing the assertion: first-order upwind's own O(dx) numerical diffusion smooths any field over the distance it travels regardless of whether the wrap is correct, so "matches exactly" is the wrong claim even for a correct wrap. A real wrap's own round-trip error drops ~62% over a 4x mesh refinement; a mirrored/clamped wrap (a throwaway mutation, built and run specifically to check this) drops only ~16% and stays several times larger throughout.
• Builds Stage 4's own golden demo, Passive Scalar Transport — the first pyflow run that steps a real simulation forward live, one timestep per rendered frame, via a new SimulationConfig section and bootstrap.py's _add_passive_scalar_transport wired through RenderWindow.run(on_frame=...). The demo's own centroid-displacement tolerance was likewise measured from a real run (~4% agreement with the closed-form prediction), not guessed.
• Full Blast Radius documentation sweep: docs/planning/roadmap.md (TASK-030's own entry, Stage 4's closing status table), docs/architecture/{sequences,icds,engine}.md, adr/ADR-003, docs/implementation/golden-demos.md, and every touched CLAUDE.md — including correcting two now-stale claims found mid-sweep (src/pyflow/engine/CLAUDE.md had said a periodic face raises UnconfiguredBoundaryFaceError, which TASK-030 makes untrue) and one honestly-flagged still-open gap (docs/planning/backlog.md: periodic boundaries now exist, but total-quantity conservation under a periodic wrap specifically is still unchecked).

Test plan

• Strict TDD throughout: wrapped_neighbour_cell's own tests, periodic_boundary.feature's three scenarios, and the golden demo's own scenario all written and confirmed red before the corresponding production code.
• Mutation testing: wrong wrap (mirrored to owner), wrong periodic distance (un-doubled), and a frozen-simulation-loop mutation each confirmed to fail the scenario meant to catch them, then reverted via targeted edits.
• make ci green locally: 603 tests, 53 Gherkin scenarios (all bound, make check-scenarios), 99% coverage, all doc-consistency checks (check-references/check-manifest/check-inventory/check-dependency-tree/check-docs/check-docs-index/check-graph/check-status) passing.
• Verified visually, not only by the regression test: the demo rendered offscreen at increasing frame counts shows the blob translating downstream and, by one full domain width of travel, wrapping around the periodic boundary.
• Real CI run on this PR — Stage 4 Completion Criterion 9 is left honestly pending in the roadmap's own status table until this run completes; will update that row once it does.

🤖 Generated with Claude Code