Always add paired fail-closed regression tests

When changing behavior to bypass a dependency or discovery step for a narrow special case, add regression coverage for both sides: prove the special case succeeds without the dependency and prove ordinary/non-special-case inputs still invoke it and propagate failures. Do not rely only on helper-level invalid-output tests; include a public workflow-level test using the relevant non-root input and a throwing dependency. Treat missing paired coverage as a blocking issue before merge, then rerun focused tests, lint, and a final diff audit.