Allowing more LLM models for Langchain::Assistance

[mengqing]

Currently it only supports OpenAI, and I can see that there's a PR for supporting Gemini #513

Is it better to make it more generic and make it extendable? Eg: be able to just extend a new LLM model without hard coding if conditions in the base class.

A lot of other LLM uses OpenAI standard specifications

[andreibondarev]

@mengqing Yes, I agree with you. It's best kept to be as generic as possible. Do you have any specific ideas/changes to propose? The current Gemini PR is a rough draft before I've thought about any abstractions at all. I've been thinking about maybe implementing a base/parent class and child classes that override specific methods, constants, etc.

[andreibondarev]

@mengqing Do you have any thoughts on how it's been implemented thus far in #513?