Added automatic json extraction from the response

[alicja-raczkowska]

Feature Description

If output_data_model_class is provided and the LLM returns e.g. "Sure! This is the response {...}", the JSON should be extracted and parsed.

Added

Added JSON extraction from within the response text.

[riccardo-alle]

@alicja-raczkowska I believe somewhere in the documentation we are saying that Llama/Mistral/Gemma don't support the output_dataclass argument because so far this feature you implemented was missing and we were relying only on pydantic <--> langchain integration. In that case, the documentation should be updated.

[alicja-raczkowska]

@alicja-raczkowska I believe somewhere in the documentation we are saying that Llama/Mistral/Gemma don't support the output_dataclass argument because so far this feature you implemented was missing and we were relying only on pydantic <--> langchain integration. In that case, the documentation should be updated.

I modified the docs to address the format instruction changes.