Give multi-stage jobs a fallback error sink

When a multi-stage scheduled job converts a caught stage exception into a generic HTTP 500, preserve the exception outside the same telemetry sink that may fail. Return a safe error code or correlation ID, and have an independent fallback logger retain stage, error class, and request or run ID.

Otherwise a reproducible application-level failure may be localized to a stage but remain impossible to root-cause after the fact. An unrelated warning present in retained logs is not evidence of causation.