feat(interval): pin PNT migration inventory

[kim-em]

Summary

• pin the PNT+ source, LeanCert/Mathlib revisions, toolchain, and all tracked Lean-source content
• inventory executable and textual interval tactics, direct imports, reviewed interface ownership, trust-residue sites, and generated Table 10/Table 12/FKS2 workloads
• add an offline integrity gate plus maintainer refresh, source-verification, inspection, and evidence-classification workflows
• record a concrete HexInterval migration acceptance corpus without promising LeanCert API compatibility

Audit surface

• 280 executable interval_decide sites and 60 executable interval_auto sites
• 16 direct LeanCert imports and six load-bearing interface families
• 107 qualified reference occurrences retained as inventory-only evidence under reviewed interface provenance
• 83 executable textual native_decide trust-audit candidates
• Table 10: 87 target sites and 38 supporting a₂ sites
• Table 12: 26 structurally parsed rows × five reviewed columns = 130 checks
• FKS2: 13,590 cells across fourteen shards
• 449 migration records intentionally remain pending; this PR establishes the audited backlog, not the migration claim

Verification

• python3 -m unittest scripts.maintenance.test_pnt_inventory (23 tests)
• offline pnt_inventory.py --check
• exact-source --verify-source against PNT+ 21998bb6196b56789f72a52656a781a75e134eb0
• --inspect-source report validated
• --require-classified rejects the expected 449 pending records
• copyright, line-count, DAG, trust-surface, Phase 4, conformance registration, factor freshness, and diff checks pass
• fresh exact-head Opus review: merge-ready, no blockers

No Lean sources or proof trust mechanisms change in this PR.