Add support for Local LLMs via Ollama

[chand1012]

In theory this should be possible with nothing more than the existing API, as Ollama supports an OpenAI-compatible API. However, there should be stricter token limits, which means smaller chunks for embeddings and a slew of other changes. Different models should be used for different use cases as well. For generating code, we should use starcoder2 or other similar model. For chat and text completions, Mistral 7b would be better. This should also be configurable by the user.