Update test.py profiling

[glenn-jocher]

Improved test.py profiling to match the new YOLOv5 PyTorch Hub profiling format:

Speed: 0.2ms pre-process, 9.2ms inference, 0.8ms NMS per image at shape (32, 3, 640, 640)

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Refined half-precision utilization and timing metrics in testing routines.

📊 Key Changes

• Renamed half_precision argument to half for clarity.
• Refactored timing code to include a pre-processing step.
• Removed an unnecessary random library import in plots.py.
• Simplified imports of an example file in plots.py for consistency.
• Adjusted train.py to directly use opt values instead of reassigning them redundantly.
• Included additional condition checks for opt.weights being a list before iterating.

🎯 Purpose & Impact

• ✅ Clarity: The argument rename makes the half-precision flag usage clearer.
• 🚀 Enhanced Metrics: New timing for the pre-processing stage helps users better understand where time is spent during inference.
• 🧹 Code Cleanup: Removal of extraneous code contributes to cleaner, more maintainable codebase.
• 👁️ User Insights: Updates provide more granular timing details, offering users more insight into their model's performance. Users interested in optimization will now have more detailed information on their model's speed with the added pre-processing time reporting.