fix: format history as string to prevent Llama 3 hallucinating assistant role during contextualization

fix: handle 2D numpy array from multilingual reranker

fix: dockerignore and memory init

chore: remove IS_RENDER and gemini imports

feat(rag): enable rerank model on render

feat(rag): use bkai huggingface model universally

fix(rag): update prompt and reduce temperature to prevent hallucinations

fix: make heavy ML libraries import conditionally on Render to prevent OOM/timeouts

feat: tích hợp supabase database và cấu hình github actions ci

“…I”

hoàn thiện Multimodal RAG chatbot với FastAPI Backend & Web UI