Write a 'Bring your own model' guide with per-backend recipes

[stefanthearchitect]

Context

The README covers BYO at a high level, but a dedicated guide with copy-paste recipes per backend would lower the barrier a lot — this is the feature that brings the local-LLM crowd in.

The task

Add docs/byo-models.md (linked from the README) with a short, tested recipe for each of:

• Ollama (local, no key)
• LM Studio (local server)
• llama.cpp server (custom provider + base_url)
• OpenRouter (key via BORG_API_KEY / api_key_env)
• OpenAI

Each recipe: the exact env vars / settings set commands, a note on picking a tool-calling-capable model, and the BORG_CONTEXT / BORG_TTFB knobs where relevant.

Acceptance

• Every recipe actually works when followed (test the ones you can).
• Cross-links to the tested-models table and the permission/auto-approve docs.