fix(storage): recluster endless loop

[zhyass]

I hereby agree to the terms of the CLA available at: https://docs.databend.com/dev/policies/cla/

Summary

This PR addresses the issue of potential fragmentation during the recluster process, preventing the system from entering an infinite recluster loop.
By optimizing the reclustering logic, this update minimizes the likelihood of producing excessively small fragments, enhancing system stability and effectively preventing infinite loop issues.

Tests

• Unit Test
• Logic Test
• Benchmark Test
• No Test - Explain why

Type of change

• Bug Fix (non-breaking change which fixes an issue)
• New Feature (non-breaking change which adds functionality)
• Breaking Change (fix or feature that could cause existing functionality not to work as expected)
• Documentation Update
• Refactoring
• Performance Improvement
• Other (please describe):

---

This change is