Added support for using local LLMs via ollama

[meuleman]

This is basic support, enabling a few embedding models as well as Llama-3.2 for chat.
The tokenizer used is currently the cl100k_base encoding by way of tiktoken.

[meuleman]

Users will need to have installed ollama, and have pulled down the model(s) they'd like to use.
It might be possible to do this automatically, and even make the list of models available a function of what is available in ollama.