Fix TensorRT exports to ONNX opset 12

[glenn-jocher]

Eliminate opset 13 warning messages:

Signed-off-by: Glenn Jocher [email protected]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Update to ONNX opset version and logging improvements in TensorRT engine export.

📊 Key Changes

• Changed the opset version for ONNX export from 13 to 12 during the TensorRT engine export process.
• Streamlined the logging format for better clarity when displaying input and output names, shapes, and data types.
• Adjusted the logging message for the FP16/FP32 engine build process to indicate the output file name.

🎯 Purpose & Impact

• Exporting models using ONNX opset 12 may increase compatibility with different versions of TensorRT, potentially benefiting users working with a specific version that has better support for opset 12.
• Improved logging contributes to a cleaner and more straightforward output, which enhances the user's ability to understand the export process and debug if necessary.
• Amended build log provides immediate and clear confirmation of what type of engine was built (FP16 or FP32) and the specific file it was saved as, aiding users in managing their model files more efficiently.