Add Groq fallback for rate limits

Fix OOM: lazy-load ML models, rename to VectorMind

feat: add frontend and backend code for multimodal RAG knowledge assistant