Fix empty-content crash with reasoning models (GLM-5.2 default)

Add PYTHONUNBUFFERED=1 so startup logs flush promptly

Load RAG index eagerly at container startup instead of on first request

Load RAG index eagerly at container startup instead of on first request

Fix: remove rag_index COPY (downloaded at runtime instead)

Initial deploy: full app with 4-tier keyword research, per-task models; RAG index served from a separate private dataset repo

initial commit