Update CoreML model conversion to do pixel scaling as well

[dlawrences]

Hi

As agreed in #133 , in order to do proper inference with a model converted to CoreML, one has to also pass pixel scaling (if there is any done).

In detect.py, there is this specified:

img /= 255.0  # 0 - 255 to 0.0 - 1.0

I have updated the CoreML model conversion step to do this as well.

Cheers

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improved CoreML model export with scaling in Ultralytics YOLOv5.

📊 Key Changes

• Updated the CoreML export process to apply pixel scaling during conversion.
• Explicitly set the image input scale to 1/255.0 with a zero bias.

🎯 Purpose & Impact

• Ensures the CoreML model processes input images with the same scaling as used in YOLOv5's detect.py.
• 📈 This change leads to more consistent performance between CoreML models and their original YOLOv5 counterparts.
• Helps users deploy models to iOS devices with correct preprocessing parameters automatically set.

[glenn-jocher]

@dlawrences great, thanks for spotting this!