- Use targeted fetch() for chunk hashes to avoid namespace cross-contamination
- Use range() scan for stale chunk ID detection
- Reduce Upstash batch size from 100 to 90