Add a content type for completion & chat responses that directly contains the structured output produced by LLMs, instead of having it embedded in a text content item as JSON serialised into a string