Simplify val.py benchmark mode with speed mode

[glenn-jocher]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Optimization of validation task settings for benchmarking speed in YOLOv5.

📊 Key Changes

• Renamed the task from benchmark to speed in benchmark.py for both segmentation and detection validation functions.
• Simplified padding and rectangle inference conditions in segment/val.py and val.py by consolidating the operations into a single line.

🎯 Purpose & Impact

• 👍 The renaming of the task parameter clarifies the intent of the code, focusing specifically on speed benchmarking.
• 🛠 The code simplification helps to make the logic behind padding and shape inference easier to understand and maintain.
• 🚀 These changes may slightly improve the performance of the validation during speed benchmarks and provide more transparent metrics for users assessing model speed.