Fix Z.ai model id, surface LLM errors as SSE frames, cache-friendly Docker layers

Dockerfile, static serving, and DEPLOY.md handoff (step 4 of deploy plan)