Add a "Last Retrained" metadata status endpoint to the ONNX ML pipeline

[Edge-Explorer]

Currently, the FastAPI backend downloads and loads model.onnx from Hugging Face on startup. We want to expose a new endpoint /api/v1/ml/metadata that returns the metadata properties stored inside the ONNX model structure (e.g., training date, feature lists, and win-rate/accuracy scores).
This will allow the frontend to display actual model performance metrics directly to the user instead of hardcoded indicators.

Requirements

1. Create Endpoint:
   • Expose a new GET /api/v1/ml/metadata route in the FastAPI backend.
2. Retrieve ONNX Metadata:
   • Access the loaded ONNX session inside backend/app/services/ml.py.
   • Read custom properties using:

     meta = session.get_modelmeta()
     custom_meta = meta.custom_metadata_map

3. API Response:
   • Return a JSON payload containing the metadata details, such as:

     {
       "trained_date": "2026-07-15",
       "features": ["rsi_14", "macd", "ema_ratio"],
       "accuracy": 0.68
     }

Code Pointers

• Backend routes: check backend/app/api/v1/
• ONNX Model loading service: check backend/app/services/ml.py

How to run locally

Refer to CONTRIBUTING.md for local environment setup with uv.