Updated cache v0.2 with hashlib

[glenn-jocher]

Possible fix for #3349

This PR increments the cache file version to 0.2 and uses a new hashlib-based solution which detects changes in dataset contents and location, recaching on any changes in either.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improved dataset hashing for better cache validation.

📊 Key Changes

• Modified the get_hash function to compute hashes based on paths (files or directories) instead of just file sizes.
• Hash now includes both the cumulative size and the actual paths of the dataset for a unique identifier.
• Removed check for 'version' in cache; now solely relies on the new hashing method to validate the cache.

🎯 Purpose & Impact

• Enhanced Accuracy: The new hashing method provides a more robust way to detect changes in the dataset, reducing the risk of using outdated cache entries.
• Increased Reliability: The combination of size and path in the hash helps prevent false cache invalidation, ensuring that cache is only rebuilt when necessary.
• User Experience: Users may observe faster setup times for repeat training sessions due to fewer unnecessary cache rebuilds.

[glenn-jocher]

Verified dataset renaming/moving incurs hash change and recaching op: