LangChain needs to improve its type system

[AndreuCodina]

I've created this LLM call to show the big problem LangChain has: https://github.com/AndreuCodina/langchain-type-checking/blob/main/main.py

Using LangChain I'd expected not to receive errors from Pyright or Ruff, or not to have to annotate myself the variable types, but it's not the case. Instead, I have to:

• Add # type: ignore[reportUnknownVariableType] twice.
• Annotate the variable types (note: they're not well-documented) and use if TYPE_CHECKING.

I really love LangChain, but basic usage feels very complicated.

Regarding the documentation, it's plenty of examples (e.g., passing a list of messages) that don't pass the type checker.
About the types like temperature, I think they should be StrictFloat instead of float.

Note: I'm using Pydantic models instead of hardcoding the keys. It's another thing LangChain should improve.