audit(bz): restore phase attestations

[kim-em]

Closes #9793.

Summary

• refresh clean scientific evidence for all fourteen remediated factor/product registrations
• close the adjacent adversarial-family gap with independently calibrated mode-3 budgets and a single clean 22-registration acceptance export
• remove the stale HexHensel dependency blocker now that every direct dependency is through Phase 4
• re-attest the preserved Phase 5–7 evidence and restore HexBerlekampZassenhaus to done_through: 7

The mode audit remains honest: the affected registrations select mode 3, use independently fixed operation-specific budgets, and explicitly give up asymptotic regression detection.

Verification

• lake exe hexbz_bench verify (22/22)
• clean verified-idle-core scientific run for all 22 mode-3 registrations
• fresh 107-case fixture emission matches the committed fixture; 55/55 dispatch traces pass
• python3 scripts/check_phase4.py
• python3 scripts/check_phase7.py
• python3 scripts/check_dag.py
• trust-surface, released-manifest, manual-split, and factor-sweep freshness checks
• lake build HexBerlekampZassenhaus HexFactorizationModules HexConformance HexManual HexBerlekampZassenhaus.FactorTacticTests (10,753 jobs)
• git diff --check

Independent review

Claude Opus identified that the fixed HO-2 adversarial ladder was advertised as performance coverage without selected-mode budgets or a Verdicts measurement. This revision budgets and measures those eight registrations, adds the missing traceability command, compares the Phase-6 baseline, and expands the Phase 5–7 re-attestation.