🐳 Fix docker compose runtime blockers on clean clone

[GiorgosPanagopoulos]

Summary

docker compose up did not work on a clean clone. Seven separate blockers, fixed together since they compound (fixing one just exposes the next):

1. Missing env_file in compose — backend service had no env_file, and the environment: block substituted ${ANTHROPIC_API_KEY}/${OPENAI_API_KEY} from the (unset) host shell as empty strings, shadowing backend/.env and tripping the must_not_be_empty validator in config.py before the app could start.
2. No automatic seeding — backend/data/seed.py was a standalone script; a clean docker compose up landed on an empty database. Refactored into an idempotent seed_if_empty(db) (returns early if suppliers is non-empty) and called it from the FastAPI lifespan.
3. Superuser seed race — the get_user_by_email → create_user sequence raised DuplicateKeyError on repeated or concurrent boots against the unique {email:1} index. Replaced with a single atomic update_one({...}, {"$setOnInsert": {...}}, upsert=True).
4. LANGCHAIN_TRACING_V2=false in backend/.env.example (was true with no API key set, which is a broken default).
5. README seed command corrected — Quick Start referenced python backend/data/seed.py (wrong path, ModuleNotFoundError). Now states sample data is seeded automatically on first startup, with the correct manual re-seed command (cd backend && PYTHONPATH=. python data/seed.py).
6. Dockerfile.backend PATH fix — pip install --target places console scripts (including uvicorn) under site-packages/bin, which was never added to PATH in the final image. The backend container failed to start on every clean build with exec: "uvicorn": executable file not found in $PATH. Found this while verifying the other five fixes end-to-end — without it docker compose up still didn't work.
7. Added .dockerignore — none existed, so every build sent the full repo (.git, node_modules, chroma_db, etc.) as build context. Confirmed 104MB and still climbing after 11+ minutes on a real npm install'd checkout, never finishing.

Also migrated the remaining Pydantic v2 .dict() calls in seed.py to .model_dump().

Verification

• docker compose config | grep -i anthropic — key resolves from backend/.env, never appears empty
• docker compose up --build — all three containers (backend, frontend, mongo) start cleanly
• curl localhost:8000/suppliers (authenticated) — 12 suppliers, 16 bids seeded
• docker compose restart backend — clean second boot: logs superuser_exists, no seed re-run, no crash
• cd backend && pytest — 160 passed
• Backend build/context transfer: 11+ minutes (104MB, still growing, never completed) before the .dockerignore fix → ~6 seconds after

Test plan

• docker compose up --build succeeds on a clean clone
• /suppliers returns 12 records after login
• Second boot (docker compose restart backend) does not crash or re-seed
• pytest — 160/160 passing