Merged UK spine: final age-tail breaks all six stored NHS identities

[MaxGhenis]

The UK spine merged in #747 runs etb_services before the final age_tail
stage. ETB derives and persists six NHS visit/spending columns from stage-time
age (packages/microcosm-build/src/microcosm/build/uk_runtime/etb_services.py:117-139,426-436);
age_tail later rewrites top-coded person.age
(packages/microcosm-build/src/microcosm/build/uk_runtime/age_tail.py:151-196);
E6 recomputes NHS from final age and requires equality with the stored values
(tools/verify_uk_identity_stability.py:378-386,431-464).

Fresh reproduction on the merged executable tree, which is byte-identical at
current main: of 400 top-coded people, 189 moved to age 85+ and 67 to age 90+.
identical_under_permutation remained true, but matches_stored_columns was
false and all six NHS columns appeared in stored_column_mismatches.

Acceptance criteria:

• Make final age and every stored age-derived NHS value obey one coherent
  production contract.
• Add an age-tail × NHS × E6 integration regression that fails on the merged
  ordering.
• Re-run E6 on the repaired final-stage artifact and attach an identity-bound
  receipt before relying on or promoting the merged spine.

---

From the post-merge verification of #747 (round-1 review + fresh reproductions on merged main); full evidence: https://github.com/PolicyEngine/microcosm/blob/review/pr-747-audit/REVIEW-747.md

🤖 Filed by the microcosm owner session with Claude Code