Haystack 3.1.1 stops a discarded malformed tool call from killing the next LLM call.

Released September 3, the OpenAI Chat Completions and Responses converters were raising on an assistant message with no content parts, which is exactly what a Chat Generator returns when it discards a malformed tool call. The message is now sent with empty content, which both APIs accept, so the run continues. ChatMessage.from_openai_dict_format accepts the same so it round-trips, and the Hugging Face converter got the matching fix. This is the failure where one bad tool call ends an otherwise healthy multi-turn run.