Update metrics.py with IoU protected divides

[glenn-jocher]

Resolve #8549

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improved numerical stability in bounding box operations.

📊 Key Changes

• Added an eps (epsilon) parameter to the box_iou, bbox_ioa, and wh_iou functions to prevent division by zero errors.
• The default value for eps is set to 1e-7 for all three functions.

🎯 Purpose & Impact

• 🎯 Prevents Division by Zero: The addition of an epsilon term in the denominator ensures there’s no division by zero, which can occur if bounding boxes have zero area during calculations.
• 📈 Increases Numeric Stability: Small denominators can cause numeric instability in computations, so introducing eps helps maintain reliable results.
• 🔍 Enhances General Robustness: By making these functions more robust, users can expect more consistent and error-free performance when dealing with a variety of different sized bounding boxes.