Upgrade langchain version to >0.0.239

[michaelchia]

Problem

I want to be able to add ChatVertexAI as a provider but it has been broken in langchain till fixed in langchain v0.0.239.
I have to do some hacky monkey patching to langchain in v0.0.220 to get it to work.

I would prefer for jupyter-ai to support langchain>=0.0.239 so I can productionalise it without ugly hacks that may cause issues.

Proposed Solution

Currently the only part I found incompatible with langchain v0.0.239 is in LearnChatHandler. In v0.0.239, langchain.schema.BaseRetriever (which LearnChatHandler inherits) was changed to be a pydantic.BaseModel. Would need a bit of refactoring to make it compatible.

Additional context

I am using jupyter-ai in a corporate setting where I add custom providers, one of which is ChatVertexAI, by adding to the "jupyter_ai.model_providers" entrypoint.

[JasonWeill]

Per #264 (same reporter), ChatVertexAI also requires the ChatPromptTemplate to be customizable.

[michaelchia]

For that at least as a temporary fix I am able to just override the _generate method in the Provider to remove that empty AIMessage before passing it to the parent class. So in terms of prioritisation, having the langchain version is much more of an issue to me.

[michaelchia]

newer versions of the anthropic sdk (>=0.3) require later version of langchain as well.

[JasonWeill]

As of Langchain 0.0.267, Pydantic 2 is now supported; see langchain-ai/langchain#6841. This provides another incentive to upgrade.