Fix APIConnectionError

✅ Solution

APIConnectionError usually arises from network issues or temporary unavailability of the hosted LLM endpoint. Fix this by implementing retry mechanisms with exponential backoff and failover logic to redundant deployments within litellm, ensuring robust error handling for temporary API outages. Specifically, add `num_retries` kwargs to embedding model groups to trigger automatic retries and properly handle `APIConnectionError` in cooldown handlers to allow failover to healthy deployments.