Improve documentation around local model configuration

[johnojo175-dev]

I've been looking through the project and noticed that the model configuration could use a bit more explanation. Since Atomic Agent is built around local inference, users will likely try different GGUF models, quantizations, and context sizes.

It would be helpful to include recommendations for:

1. Supported model families
2. Recommended quantizations for different hardware
3. Memory/VRAM expectations
4. Common configuration pitfalls

I think this would reduce setup friction, especially for people who are new to running local LLMs.

[sosidudku1]

Thanks, this one points at a real gap. A local-model setup guide (supported model families, quantization vs RAM and VRAM, common pitfalls) is planned as part of the Quick Start docs effort. PR #38 will also make model setup itself much simpler. Closing here; the guide will land in the docs.