fix(sim): transfer subject/vassal bonds during empire merges

[csheldrick]

Summary

• mergeEmpires transferred fleets, artifacts, factions, alliances, and diplomatic relationships from the absorbed empire to the dominant one, but never touched state.subjects. A merged empire that was a vassal/tributary (or an overlord of others) left dangling subject records — the next stepSubjects pass would silently drop the relation or fire a spurious liberation event, instead of correctly carrying the bond forward.
• Subject/vassal ties now transfer to the dominant empire on merge, mirroring the cleanup removeEmpireFromGalaxy already performs on empire death. A tie directly between the absorbed and dominant empire collapses (both sides are now the same empire); a transfer that would violate the one-overlord/no-cycle invariant falls back to liberation, same as elsewhere in the codebase.
• Declared npm run build / npm run lint / npm run report as checks in .noet/VERIFICATION.md and recorded passing evidence via noet verify run + noet verify check (includes the determinism replay guard).
• Recorded one constraint in .noet/CONSTRAINTS.md: new per-empire relational state needs transfer logic in both mergeEmpires and removeEmpireFromGalaxy, since this bug is exactly a case of the former being added without touching the latter's counterpart.

Test plan

• npm run build (typecheck + Vite build)
• npm run lint
• npm run report -- --milestones 1000,3000,10000 — 10k-tick run with 20 empire merges and ongoing subject churn, no crashes; determinism replay check passed
• noet verify run / noet verify check — all declared checks green and fresh

🤖 Generated with Claude Code

---

Generated by Claude Code

[csheldrick]

Two changes are needed before I would merge this.

1. Avoid the new Merge.ts ↔ Subjects.ts circular dependency. Subjects.ts already imports mergeEmpires; this PR makes Merge.ts import breakSubjectRelation, subjectOf, and subjectsOf back from Subjects.ts. Even if the current ESM cycle happens to initialize successfully, it creates a fragile dependency between core simulation modules. Move the shared subject-query/transfer primitives into a neutral module, or implement the merge-side state transformation without importing the module that calls mergeEmpires.

2. Add focused regression coverage for the subject-transfer branches. The 10k-tick report is useful integration evidence, but it does not prove the specific cases introduced here. Cover at least: absorbed subject transfers to dominant; absorbed overlord's subjects transfer to dominant; a direct dominant↔absorbed subject tie collapses; conflicting transfer liberates instead of creating a second overlord or chain; no surviving relation references the absorbed ID.

The Noet record added by this PR is genuinely useful and appropriately captures the cross-cutting invariant. The implementation direction also looks correct; these requests are about making the fix structurally safe and checkable.

[csheldrick]

Addressed both points:

1. Extracted the subject-relation primitives (subjectOf, subjectsOf, isSubjectPair, createSubjectRelation, breakSubjectRelation, plus the status label/profile tables) into a new src/sim/SubjectRelations.ts, which has no dependency on Merge.ts. Merge.ts now imports from there instead of Subjects.ts, breaking the cycle (Subjects.ts still imports mergeEmpires for peaceful integration, and re-exports the primitives so existing call sites are unaffected).
2. Added src/sim/Merge.test.ts (Node's built-in test runner, no new dependency) covering exactly the branches you listed: absorbed's subject status transfers to dominant, absorbed's own subjects transfer to dominant, a direct dominant↔absorbed tie collapses, and conflicting transfers on both the subject-side and overlord-side liberate instead of creating a chain/second overlord — each asserts no surviving relation references the absorbed empire id. I confirmed these tests fail against the pre-fix code and pass against the fix. Wired up npm run test:unit + a CI step for it.

All declared noet verify checks (build, lint, test:unit, headless report) are green.

---

Generated by Claude Code