Add support for the Chat input for the TGI model

This does not seem possible for the moment as the `InferenceClient` class of `huggingface_hub` has 2 different methods: `generate_text` and `chat_completion`. Only the 1st one supports structured output while only the 2nd one supports the chat input.