Update trainloader in late training

[glenn-jocher]

This PR allows for trainloader updating midway through training. In partial fulfillment of feature request #4461

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improvement to the training data loader during the late training phase in the YOLOv5 model.

📊 Key Changes

• Conditional logic added to update the training data loader at 90% of the training epochs.
• A new data loader creation call during the update step with revised parameters for batching, augmentation, and worker processes.

🎯 Purpose & Impact

• 🔄 The purpose of this change is to refresh the training data loader towards the end of the training process, which may help the model to further improve by seeing the data in a slightly different way or more efficiently.
• ✅ The impact on users could be improved model accuracy and potentially more efficient training in the final epochs, without affecting earlier stages of the training process.

[github-actions]

This pull request has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions YOLOv5 🚀 and Vision AI ⭐.