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

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