Add local models, supervisor process

[matildepark]

If a user doesn't even know how to use a model or anything, we should help custody and supervise an Ollama process and use it as a fallback option. There's a few projects that already do this. Function calling may be an issue. Worth investigating. Likely intersects with moving off llm-api (#16, #5).

[matildepark]

See https://github.com/hdresearch/nolita/blob/ty/replace-zod-gpt/src/agent/llamaCompletion.ts

[AlephNotation]

closed by #124