RagPipeline.load() now falls back to downloading faiss.index and
chunks.parquet from the whosouravsharma/medical-assistant-artifacts
dataset when they're not present locally, so the Space doesn't need
the binary artifacts committed to its own git history.
Replace default Streamlit template with medical RAG chat UI
Wires src/streamlit_app.py to RagPipeline, adds the rag package under
src/, and updates the Dockerfile/requirements for the pipeline's deps.
artifacts/ is a placeholder for now — index build is a follow-up.
Duplicate from streamlit/streamlit-template-space