Finalize deployment: local model only and fixed sheets source

Docker

Use n8n-first price fallback chain and show prediction confidence

Add /health endpoint for Render readiness checks

Lazy-load model after server startup to avoid port bind failures

Use gdown for Google Drive model download fallback

Fix Drive model download flow and PyTorch 2.6 checkpoint loading

Add fallback model URL for Render startup

Remove hardcoded Flask host and port in local run block

Support external MODEL_URL download for checkpoint on startup

Initial project commit for deployment