Add a way to view unformatted response when using inference

[ashduino101]

Currently, in the frontend, the assistant response is formatted no matter what. If the assistant responds with malformed or incomplete formatting (e.g. three backticks in multiple locations that aren't meant to represent a code block), the response may render incorrectly. It would be nice to be able to view the plaintext (unformatted) response from the assistant in cases like this.