Stream chat tibble

[JamesHWade]

• Stream chat now returns a chat_tibble class (stream() should eventually return the same structure as chat()  #21).
• Fixed a bug with chat() trying to create a tibble with empty tool_calls column.

[JamesHWade]

Implemented a ... input from chat() and stream(). This sticks with a chat_tibble output, which is different from #23. Not sure which you prefer @romainfrancois.

[romainfrancois]

I think having ... in the main function is not that great after all. I'll rework #23 so that chat() takes a single argument called messages as in the API, and then we can have a set of building functions to make messages.

It's a little bit less magical ✨, but I think that's the right balance.

I also think I agree with you that the tibble output was the right call. One of the goals of #23 was to not lose the inputs, but perhaps we can organize the tibble so that it has both inputs and outputs.

msgs <- chat("you are a weather reporter. in the following messages, I'll give you a city and you give back a weather report. Ready ?")
chat(messages(msgs, "Montpellier, France"))
chat(messages(msgs, "San Francisco"))

It's important here that we retain the original prompt.

I'll play some more on #23

[romainfrancois]

I changed a few things in #26 I can take the spirit of this PR and adjust it

[JamesHWade]

Okay, I'll close this one then. Happy to see the continued progress! It's shaping up nicely.