OpenAI now splits rate-limit errors: 429 'slow_down' vs 503 overload

OpenAI's API now returns a `429` with the `slow_down` code when your traffic is ramping too fast, and a `503` with `server_is_overloaded` when the model itself is saturated. Branch your retry logic on the code: wait at least the `Retry-After` header when it's present, exponential backoff when it isn't. A blind "retry on any error" loop now punishes you in the one case — `slow_down` — where you should be easing off concurrency instead.