Assert --optimize not used with cuda device

[democat3457]

TorchScript throws a lot of errors complaining about mobile-optimized models not working with CUDA devices when trying to use a mobile-optimized cuda-export-device model, regardless of whether the model is ran on cpu or a cuda device. This PR prevents that from occurring by ensuring that --optimize cannot be used with a cuda export device.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improve export script by enforcing CPU device usage during optimization.

📊 Key Changes

• Added a condition to assert that the --optimize flag requires --device cpu.

🎯 Purpose & Impact

• 🎯 Purpose: Ensures users avoid compatibility issues when using the --optimize option by running the model optimization on CPU only.
• 💡 Impact: Prevents potential errors for those attempting to optimize models with CUDA-enabled devices, leading to a smoother user experience when optimizing YOLOv5 models for deployment.

[glenn-jocher]

@democat3457 PR is merged. Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐