[Feature Request] GGUF export and usability improvements for running locally

[scosman]

I think we should make going from cloud to local and back super user friendly. Currently you can download cloud-fine-tunes now, but need to use some libraries/CLI to get them running locally.

This is a WIP spec - please comment below on what you'd want in this space!

• Download GGUF format from the "Fine-tune" tab (just Lora or flattened model?)
• Automatic Ollama integration: One button to export model to Ollama?
• Is Ollama enough, or do people want llama.cpp, vLLM, and other engines?

Priority here is Fireworks since they have 60+ downloadable models compared to Together's 4. Downloading is easy (https://docs.fireworks.ai/api-reference/get-model-download-endpoint) but we'll need to unpack, convert and maybe quantize.