Feature Request: Better error transparency from LLM response

[zarlor]

Currently no matter what the issue an LLM might have with a request Danswer seems to generally respond with simply "LLM failed to respond, have you set your API key?" when it can be easily verified, even by starting another chat session with the same assistant, that, in fact, the API key is set and everything would otherwise seem to work fine. I THINK what is happening is the LLM is returning some error message and that is Danswer's standard response back to the user. It would be nice to be able to see the actual response sent by the LLM to better troubleshoot what the error might be (I think most likely the situations I've run into with this are just context is full for the overall chat session, but I'm not positive if that is the case or not, hence this request).