A part that can appear in a response received from the LLM (e.g. text, reasoning, or tool calls).
Optional cache-control directive for the provider's prompt-caching feature.