FN-5696: backfill missing mission feature assertion links

Backfill and repair missing mission feature assertion links so validator coverage stays intact for legacy mission data.

- add MissionStore backfillFeatureAssertions with dry-run and mission-scoped repair support
- centralize feature assertion text derivation and reuse it for create/update/backfill flows
- add regression tests for repair behavior, idempotency, and dry-run semantics
- document the operator repair workflow and add a dedicated backfill script for FN-5696

Files changed:
 docs/missions.md                                  |   2 +
 packages/core/src/__tests__/mission-store.test.ts |  84 +++++++++++
 packages/core/src/mission-store.ts                | 116 ++++++++++++++-
 scripts/backfill-fn-5696-feature-assertions.mjs   | 174 ++++++++++++++++++++++
 4 files changed, 373 insertions(+), 3 deletions(-)

Fusion-Task-Id: FN-5696

Fusion-Task-Lineage: 483bf2cc-be03-4007-860b-66d841e6aa15

docs/missions.md

@@ -137,6 +137,8 @@ Fusion keeps a canonical per-feature assertion invariant in `MissionStore`:
 - This applies to all creation paths (interview import, API, CLI, tools).
 
 Assertion text source priority is: `acceptanceCriteria` → `feature.description` → fallback text (`"Verify implementation of: {feature.title}"`).
+
+**Operator repair note (FN-5696):** Some databases created before the feature-create-path fix could show feature `acceptanceCriteria`/`description` in the UI but still have zero `mission_feature_assertions` links, which caused validator auto-pass short-circuits. Run `node scripts/backfill-fn-5696-feature-assertions.mjs` to preview repairs, then `node scripts/backfill-fn-5696-feature-assertions.mjs --apply` to write links. Use `--mission=<missionId>` for scoped repair (for example the confirmed Goals mission `M-MP32KU9Y-0001-2ADN`).
 - **Verification fields**: Milestone and slice verification criteria from the interview are stored in dedicated `verification` fields rather than concatenated into descriptions
 - **Milestone acceptanceCriteria derivation**: explicit `milestone.acceptanceCriteria` from interview output is authoritative. When omitted/blank, Fusion derives a deterministic bulleted summary from child features after creation: prefer `feature.acceptanceCriteria`, fall back to `feature.description`, skip empty contributors, and leave milestone acceptance empty when nothing contributes
 - **Partial plans handled**: Auto-generation is robust to partial plans (missing slices/features or empty criteria) without throwing errors