AI microservice: Scan and anchor detection pipeline

Implement a Python FastAPI microservice responsible for:
- Accepting uploaded scan files (GLB/USDZ) via /process-scan.
- Running mock (and future real) mesh/semantics parsing to produce anchor points.
- Returning anchor metadata (id, name, position, rotation, confidence) for integration with the backend.
- Scalable architecture to swap in real analysis (RoomPlan JSON, pygltflib/trimesh) or cloud API calls.
- Health, version endpoints, basic logging.

Acceptance:
- /process-scan POST endpoint works for demo and returns detectable anchors for a scan.
- Integration with backend complete/processing lifecycle.
- Runs as a service in Docker, connected via Docker Compose.