test(workbook): qualify terminal cell persistence

[rudycelekli]

What & why

Persist terminal workbook cell outcomes before the worker-owned ORM session closes. False conditions and missing AI/research prompts now commit their existing receipt and row status after the existing run fence, preventing old COMPLETE values from surviving into later hydration.

This owned-fork PR runs the unchanged four-job qualification workflow for the candidate.

Verification

• Exact main native real worker + migrated SQLite: one success control passes and six terminal cases fail. Candidate: all seven pass.
• Six actual worker controls change the real queue owner/lease or cancel the job and pause the workbook after its preflight transaction closes. All three terminal paths reject the lost lease and preserve the old COMPLETE=42 row and receipt. No production callable stub. Independent root native replay passes all six controls.
• Expanded focused backend suite: 156 pass, no skips; selected run-once/dependency/failure API regressions: 15 pass. AST parse and diff whitespace checks pass.
• Frozen hashes, original before/after proofs and independent cold review are retained in the private campaign record. Full hosted CI is pending at creation.

Boundaries

Only terminal result durability changes. Existing condition evaluation, terminal statuses and run-state checks are preserved. No HTTP authentication, tenancy, provider authorization or model/network behavior is claimed by the local SQLite proof. Hosted PostgreSQL/frontend qualification is pending.

Checklist

• Commit carries Rudy's DCO sign-off
• Scope is one workbook result durability fix
• Independent source and native fence review
• Full existing four-job CI passed

[rudycelekli]

Superseded by owned-fork validation #7 using the identical reviewed tree with an explicit GitHub-verified SSH signature. This original head has a DCO sign-off but no cryptographic signature; its qualification receipts remain preserved. No upstream publication.