Run the aggregate checkpoint with the mechanical ownership audit

[pinin4fjords]

Parent: #1745

Predecessor: #1757 and its stacked draft PR.

Prototype evidence: draft PR #1743 and checkpoint 3f11f175. The checkpoint verification matrix and failure classification are preserved in PR #1743.

Outcome

Run the complete stacked implementation through the aggregate routing, topology, static, render, simplify, and independent #1653 gates, with a mechanical ownership audit. This issue validates and removes only what the reviews identify as dead; it introduces no new architecture.

Note: completing this checkpoint proves the #1745 ownership migration, not all of #1671. #1671 closes only after the quality-half disposition review (#1760) has resolved what remains outside this stack.

Estimate: 4-8 focused hours plus CI/render-preview time. Each behavioral child already ran $pinin4fjords:simplify on its own diff (per #1745), so the aggregate pass here is a verification sweep, not a first cleaning.

Scope

• Run the four-case gate first, spelled by test id (below).
• Run the broad routing gate, topology validator, and render corpus.
• Run the exact whole-repository static gates, hooks, and full test suite.
• Run $pinin4fjords:simplify over the aggregate production diff as a verification sweep; apply bounded reuse/dead-code findings through the sole writer.
• Run the mechanical ownership audit (below).
• Obtain independent code, visual I/N/D, and Architecture programme: semantic route planning for first-render robustness #1653 reviews.
• Push a final commit without a CI-skip marker and update the aggregate draft PR description to describe the aggregate diff against main.

The mechanical ownership audit

Every behavioral child's PR body carries a "Coordinates transferred / setters deleted in this PR" section. This issue takes the union of those lists and verifies mechanically:

• every listed setter is deleted (grep returns nothing, or only the explicitly retained entries with their cited evidence);
• no migrated coordinate has a second writer (each retained legacy pass named in Retire the pre-request convergence settlement chain and land the four-case gate #1761 either subsumed-and-deleted or retained with evidence it cannot write a granted coordinate);
• git grep -n 'finalized_owned_segments\|_scalar_component_plan\|_settle_reserved_trunk_axes\|_shared_left_exit_opening\|_landing_domains\|landing-obstacle' -- src tests prints nothing (these were checkpoint-branch-only or never existed; their absence is expected, verify rather than hunt);
• exactly one production integration call site for solve_corridor_cohorts, proven by the Trace final settlement and resolve seed-72 routing defects #1746/Join convergence variables into the one cohort problem #1752 instrumentation counters (at most two compiles per render: intent, final; exactly one final solve) together with an exact structural check over call expressions (AST-based, e.g. ast.walk filtering ast.Call nodes whose func resolves to solve_corridor_cohorts across src/). A textual git grep 'solve_corridor_cohorts' -- src is a helpful auxiliary audit but its raw count also matches imports, exports, and the definition, so it is not the proof;
• no synthetic-prefix ownership decoding, connector-direction boundary inference, or validation exemption exists;
• every production compatibility component carries typed whole-group provenance (bound at Join convergence variables into the one cohort problem #1752, plan-level at Publish and enforce the cohort-final freeze #1757); none arises from a silent incompleteness fallback, and no migrated coordinate sits under a legacy owner via compatibility;
• the stage trace shows at most one GENERAL and one APERTURE batch, at most two compiles (intent, final), exactly one final solve, and zero post-cohort-final mutation, on every corpus fixture;
• the render path's re-observation count does not exceed the frozen sequence's requirement (no regression against main's settlement round count outside the sanctioned GENERAL and APERTURE batches).

The four-case gate, spelled by test id

pytest -q -n 0 \
  "tests/test_member_geometry.py::test_seed_15_routes_split_siblings_as_independent_members" \
  "tests/test_member_geometry.py::test_seed_15_wraps_u_bypass_above_crossing_merge_trunk" \
  "tests/test_member_geometry.py::test_seed_77_shortfall_requests_one_atomic_corridor_aperture" \
  "tests/test_member_geometry.py::test_reservation_reroute_reseats_port_peeloff_after_reconciliation" \
  "tests/test_corridor_cohort_integration.py::test_opposing_bypass_lines_hold_separated_direction_qualified_lanes" \
  "tests/test_corridor_cohort_integration.py::test_no_solver_owner_contains_opposite_running_claims"

These landed together at #1761 and must all still pass, plus test_post_freeze_stages_are_validation_only from #1757. Confirm the collected count matches before believing a pass.

Non-goals

• No new planner, constraint, owner relationship, retry, clearance cycle, render stage, fixture branch, or collision-specific exception.
• No weakening or xfail of an acceptance gate.

Verification

pytest -q -n 0 tests/test_member_geometry.py tests/test_corridor_cohort_integration.py tests/test_corridor_cohorts.py tests/test_convergence_corridor_grants.py tests/test_deferred_final_guards.py
pytest -q -n 0 tests/test_routing_gate_coverage.py tests/test_guard_registry_golden.py tests/test_guard_registry.py
pytest -q -n 0 tests/test_topology_validation.py tests/test_hash_seed_determinism.py
pytest -q -n auto
ruff check src/ tests/ scripts/
ruff format --check src/ tests/ scripts/
mypy --no-incremental
prek run --all-files

Render review: the authoritative diff is the PR preview at https://seqeralabs.github.io/nf-metro/_pr/<PR_NUMBER>/, produced by .github/workflows/pr-renders.yml. Confirm the four hash_seed_determinism/seed_* entries added to scripts/gallery.yaml in #1746 are present so the seed renders appear in the diff. Classify every delta I/N/D by eye; a green suite is not evidence of no visual regression, and every changed render is eyeballed.

#1653 hard stop

Any failing gate that appears to require a new mechanism stops this issue. Record the structural evidence and open a separately reviewed design issue rather than expanding the validation slice or applying a sequential repair.