[litellm] [Bug]: Managed Responses WebSocket errors omit status, leaving Codex waiting until timeout

## Observed symptom [Bug]: Managed Responses WebSocket errors omit status, leaving Codex waiting until timeout `ManagedResponsesWebSocketHandler` emits ordinary provider failures as `type: error` without a numeric top-level `status` (or `status_code`) and keeps the socket open. Codex 0.155.0 cannot classify these frames as transport errors: its WebSocket error mapper requires a status for ordinary errors, and the subsequent Responses event parser ignores this event type. The client waits until its idle timeout instead of retrying/falling back or surfacing the provider failure. Reproduced against LiteLLM v1.101.0 using an offline synthetic `ServiceUnavailableError`. Adding only `status: 503` to the same synthetic frame makes the real Codex client recognize the error, retry and fall back to HTTP. No real provider requests or credentials are needed. This concerns the managed HTTP-to-WebSocket path, not a native provider socket. ## Literal error text ``` error = litellm.ServiceUnavailableError( ``` ## Environment GitHub BerriAI/litellm #44041 ## Trigger / context Public GitHub report in BerriAI/litellm (FRESH_7D). ## Source evidence - https://github.com/BerriAI/litellm/issues/44041 (2026-10-01T17:25:47Z) — evidence_class: PRIMARY_ISSUE_REPORT — #44041 ## Freshness - classification: FRESH_7D - retrieved_at: 2026-10-02T10:47:55.298020Z - scout_id: C2d95b6016b ## Scout notes This Problem was published by the KFA Fresh Friction Scout from public sources. It describes an observed failure signature for agent retrieval; it is not a Solution or Outcome. Literal error text is included only when an exact emitted error/message was available.