demo: simple FastAPI inference endpoint

[infinityengi]

Create a minimal FastAPI app that:

• Loads a saved model
• Accepts image upload
• Returns segmentation as PNG / overlay

Checklist

• Implement app/app.py with endpoint /predict
• Add requirements in docker/requirements.txt optional extras
• Add demo notebook showing using the API

Acceptance criteria

• App runs locally and returns a segmentation overlay for a sample image