chore(docker): package app services and slim runtime image

[X0luka]

Summary

• package FastAPI and Streamlit in Docker Compose alongside Qdrant
• use one shared rag-agent-app runtime image for api and ui services
• move RAGAS to the eval optional dependency so offline evaluation deps stay out of Docker
• make the Streamlit API base configurable with API_BASE

Validation

• uv sync --frozen --no-dev
• uv run ruff check src ui eval/run_eval.py scripts tests
• uv run pytest tests/test_chunker.py tests/test_bm25.py tests/test_hybrid.py -v
• uv run python -m json.tool eval/golden_set.json >/dev/null
• docker compose build
• docker compose up -d
• docker compose ps
• curl -sf http://localhost:6333/
• curl -sf http://localhost:8000/health
• curl -I http://localhost:8501
• docker run --rm rag-agent-app:latest python -c "import importlib.util; print(importlib.util.find_spec('ragas'))" -> None

Notes

• rag-agent-app image size after rebuild: 1.15GB
• API/UI containers are left running locally for manual testing