Security assistant chat as the front page (Nox-style, WebLLM on device, grounded in the Devseis knowledge base, Run audit from chat); phone check moved to phone.html

Model v0.3 results: 100% on held-out data, 36/36 report parts in a real Mac audit

Use fine-tuned model v0.1 (WebLLM, constrained JSON decoding, q0f16 on desktop / q4 on phones); MLC converter; v0.3 pipeline

MODEL_LLM.md: v0.3 run needs --max-len 2560 (measured)

MODEL_LLM.md: datasets and their versions

MODEL_LLM.md: fine-tuning versions, current run and next steps

Add docs/MODEL_LLM.md: role of the language model and how to improve it