Remove DDP destroy_process_group() on train end

[glenn-jocher]

May resolve #7307

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Optimized distributed training cleanup process in YOLOv5.

📊 Key Changes

• Removed redundant code that manually destroyed the process group after training.

🎯 Purpose & Impact

• 🚀 Purpose: Streamlines the distributed training process to be more efficient.
• ✨ Impact: Can lead to smoother shutdown of training processes, possibly reducing errors on exit for users utilizing distributed training environments. This change primarily affects users working with multiple GPUs or across multiple nodes.