Persist DaveAI production runtime repairs - #2186

What changed

• Persisted the DaveAI production-only runtime repairs into vps/patches/daveai-production-runtime-20260731/:
  • backend brain_llm.py native Ollama streaming/chat repair
  • Nginx vhost/rate-limit/health/provider-vault fixes
  • TD2 generated asset manifest and cache-busted production bundle
  • Windsurf/Ops auth-gated production HTML
  • idempotent apply script and LiteLLM functional health script
• Added repeatable DaveAI production E2E coverage:
  • vps/production-e2e-runner.cjs
  • npm run test:daveai:prod
  • .github/workflows/daveai-production-e2e.yml
• Added VPS maintenance diagnostic script: vps/vps-maintenance-check.sh.
• Added the production E2E audit/roadmap doc.

Validation

• git diff --cached --check
• node --check vps/production-e2e-runner.cjs
• bash -n for runtime apply, LiteLLM health, and VPS maintenance scripts
• py_compile for persisted brain_llm.py
• Public E2E runner without auth credentials: 6 passed, 0 failed, 1 skipped
  • skipped: authenticated chat reply, requires DAVEAI_E2E_EMAIL / DAVEAI_E2E_PASSWORD
• LiteLLM functional health script: readiness/liveliness/models/completion all passed; bare LiteLLM /health remains a bad/hanging probe and is documented separately.

Notes

• Local proofs/ artifacts were intentionally not committed because they are ~20 MB of screenshots/JSON. The audit doc references the local proof roots.
• VPS reboot/apt upgrade was not performed; maintenance-window steps are documented and scripted as diagnostics only.