Data distribution rebalancing sometimes stops in an unbalanced state

I've seen a few situations where a cluster showed a notable disparity in the bytes stored distribution across its storage nodes, and data movement would rebalance in order to resolve that disparity. Then for reasons I have not investigated, data movement would stop prior to parity being restored.

Bouncing the cluster seemed to get data movement rolling again, and it would make more progress before eventually stopping.

No files, tests, or entry points are named. Start by reproducing the issue with a cluster that has uneven bytes across storage nodes, then trace why data movement stops and whether restarting resumes it. Done means rebalancing reliably reaches parity without requiring a cluster restart, with coverage for the stopping case.

Written by the indexing model from the issue text.