Retain current108 CodeQL diagnostic results with compact replay

[DavidBakerEffendi]

Results

Retains the actual 108-case CodeQL run under registered plan 06, with compact evidence and CI replay. Raw observations are 55 reached, 52 not-reached, and one runner-error (MissingExactEndpoints on infeasible-branch-negative). Normal reports preserve 107 inconclusive and one runner-error because aggregate resource qualification remains unavailable. This does not activate scoring, freeze results, or publish a release.

The unchanged 150-second extraction / 75-second shared-analysis contract was used with the exact restored CodeQL 2.27.1, Swift 6.3.3, patched extractor, SDK and pack inventories. All 656 case commands exited zero with no timeouts and tracked-process cleanup. The run finished without a stop marker.

Validation

• Full compact replay rederives observations from decoded query rows and checks plan, configuration, runtime identity, command arguments, phase order, budgets, sources and exported reports.
• Five regression tests passed, including missing rows, promoted outcomes, altered decoded evidence and preserved endpoint errors.
• Replay and regressions passed from a staged-files-only copy without local databases or compiler caches, and replay passed again after rebasing.
• All 108 local database metadata, archived sources, schemas and full closure snapshots matched. Full database payloads remain local; the committed bundle is compact command/source/query evidence (~38 MiB).
• Runner/exporter focused suites: 30 tests passed before launch. Original fixture/compiler-output whitespace remains unchanged in hash-bound evidence.

Related to #220; does not close the qualification/publication issue. Parent review and merge remain separate.