Avoid FP64 ops for MPS support in train.py

[glenn-jocher]

Resolves #7878 (comment)

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Refinement of data types in weight computation functions.

📊 Key Changes

• Changed np.int to the built-in Python int for type casting in labels_to_class_weights and labels_to_image_weights.
• Ensured returned weights are of type float in labels_to_class_weights.

🎯 Purpose & Impact

• 📈 Consistency: Using Python's built-in int type may offer more consistent behavior across different platforms and versions of Python.
• 🔍 Precision: Ensuring weights are in floating-point format (float) enables more precise calculations, especially important when weights are used in loss calculations or gradients during model training.
• 🚀 User Experience: These backend changes ensure more reliable and accurate training processes for users, although most might not notice the subtle changes directly.