Fix default runtime path so Evo X2 Tailnet is primary in the web app

[terisuke]

The browser app can currently fail draft evaluation when started with the plain server defaults because the LLM client defaults to local llama.cpp at 127.0.0.1:8081. That contradicts the intended runtime order: Evo X2 Ollama over Tailscale OpenAI-compatible API first, Evo X2 llama.cpp second, workstation-local fallback last.\n\nObserved 2026-05-03:\n- GET /api/models returned HTTP 500 because http://127.0.0.1:8081/v1/models was connection refused.\n- POST /api/drafts with SSE opened the stream but emitted DRAFT_GENERATION_FAILED because http://127.0.0.1:8081/v1/chat/completions was connection refused.\n- The UI status text made the path look like Evo X2 even when the endpoint was local.\n\nAcceptance criteria:\n- Plain web-app startup defaults to the Evo X2 Tailnet OpenAI-compatible API, not workstation-local llama.cpp.\n- Fallback chain remains Evo X2 llama.cpp then workstation-local llama.cpp.\n- Phase model defaults match the current ADR: style/article lightweight Gemma, brief Qwen, draft Gemma 31B, verify lightweight Gemma.\n- UI draft status shows the actual endpoint/model reported by SSE.\n- The favicon 404 is removed or explicitly handled.

[terisuke]

Implemented and merged in PR #65. Plain web-app startup now defaults to Evo X2 Tailnet primary, keeps Evo X2 llama.cpp and workstation-local fallback order, exposes the actual SSE endpoint/model in draft status, and removes the favicon 404 noise. Validation recorded in docs/validation/runtime-ui-ddd-audit-2026-05-03.md.