Add framework for interfacing with arbitrary LLM

[caufieldjh]

Current dependency on OpenAI API means OntoGPT is set up to assume that it will make API calls to a that resource, but to use other APIs and local resources we will need to generalize our API calls.
Perhaps a modular implementation system a la OAK is appropriate.

[cmungall]

or maybe just reuse an existing modular abstraction layer. Langchain?

[caufieldjh]

Oh neat. I'm going to try that first.

[caufieldjh]

Next steps:

• Retrieve remote models if necessary - use process like OAK
• Apply GPT4All models through Langchain
• Apply other local models through Langchain
• Tests for local models
• Support non-OpenAI APIs
• Tests for non-OpenAI APIs

[caufieldjh]

Extra bonus step:

• Add support for Huggingface Hub models through Langchain (note this needs to store another API key)