Add wrapper that casts LLM response to a string.

[Zsailer]

Fixes #9

[github-actions]

👈 Launch a Binder on branch Zsailer/jupyterlab-magic-wand/chat-model

[Zsailer]

@jtpio I think this fixes the issue you saw.

[jtpio]

Nice, thanks @Zsailer!

So does it mean the response may be different depending on the provider being used?

Out of curiosity, which provider are you using for testing locally? (to check if I could also try to reproduce in my environment)

[Zsailer]

So does it mean the response may be different depending on the provider being used?

Yes, although it's really just two output types from langchain. Jupyter AI has to handle this too: https://github.com/jupyterlab/jupyter-ai/blob/7b4586ef2adc560b457731ae07be87e348a8769e/packages/jupyter-ai/jupyter_ai/chat_handlers/base.py#L454-L461

I'm checking against BaseMessage here since it's more generic than AIMessage (which is used in Jupyter AI).

Out of curiosity, which provider are you using for testing locally? (to check if I could also try to reproduce in my environment)

I'm using an internal provider :).

[jtpio]

Just tried locally using the wheel built by check-release on CI, and it works fine, thanks!

[Zsailer]

Thank you, @jtpio!