Link fuse() to AutoShape() for Hub models

[glenn-jocher]

This PR disables fusing for autoshape=False Hub models.

@AyushExel

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Update to model loading with optional layer fusion in hubconf.py and DetectMultiBackend class.

📊 Key Changes

• The DetectMultiBackend class has a new fuse parameter that controls whether model layers are fused beneficial for inference time.
• In hubconf.py, the layer fusion is now linked to the autoshape parameter when preloading a model.

🎯 Purpose & Impact

• Purpose: To give users the option to speed up model inference by fusing the layers of the neural network when the model is loaded.
• Impact:
  • 🚀 Could improve inference speed by reducing the number of operations performed during a forward pass.
  • ✅ Offers a flexible choice to users: they can turn layer fusion on or off, potentially trading off between model speed and adaptability based on their requirements.