Make cache saving optional

[glenn-jocher]

Place dataset cache-saving ops into a try: except statement to catch non-writable directory failure mode in #2976.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improved cache saving robustness for dataset labels in YOLOv5.

📊 Key Changes

• Added a try-except block around the cache label saving logic.
• Included exception handling for potential write errors.

🎯 Purpose & Impact

• Ensures smoother user experience: The change helps prevent the program from crashing if it cannot save the label cache file, informing the user instead of failing silently.
• Better error logging: If a cache cannot be written due to directory permissions, the user is alerted with a clear warning message.
• Enhanced fault tolerance: The update makes the system more robust by handling exceptions during file operations, which can occur in restricted file system environments.

[glenn-jocher]

Note that cache saving is merely useful for reducing subsequent time-to-train but has no effect on current training. On every training if a cache is not found then the dataset is scanned again, and a cache is saved for next time if the directory has write access.