Add spec ops and config expectations to deployment guide

[yellowman]

Summary

• add spec-derived ops, retention, and backup defaults to the deployment guide
• document configuration management expectations for database-managed knobs versus environment-only secrets

Testing

• python -m compileall liminallm

---

Codex Task

---

Note

Major expansion of DEPLOYMENT.md with zero-to-one checklist, backend lanes (local GPU LoRA and API), ops/safety defaults, and DB-driven config guidance.

• Docs: docs/DEPLOYMENT.md
  • Scope expansion:
    • Add zero-to-one Linux checklist and clearer prerequisites (GPU/CUDA/ROCm notes, TLS/nginx).
    • Clarify config cheatsheet (local base model path, adapter/API key vars) and writable data dirs.
  • Backend lanes:
    • Local GPU LoRA: setup, paths (/srv/liminallm/models, /srv/liminallm/adapters), JAX GPU wheel guidance, run flags, adapter modes.
    • API backend: provider keys/URLs, base_model handling, scenarios (managed only, hosted+local adapters, prompt-only), provider switching note.
    • Hybrid deployments: mix local and API; filesystem artifacts remain source of truth.
  • Model handling: base models frozen; adapters stored/streamed; training writes adapters only; request parameters (base_model, adapter_id, adapter_mode).
  • Ops & safety defaults: observability/health checks, retention windows, alerting, backup strategy, safety rails.
  • Configuration management: most runtime settings managed in DB via /admin; limit env vars to infra/bootstrap secrets.
  • Operational notes: docker/manual steps, reverse proxy pointers, troubleshooting commands.

^{Written by Cursor Bugbot for commit fe0de27. This will update automatically on new commits. Configure here.}