clean redundancy replica strategy

Now we use a flag to control wether rm the redundant replica.
We need to take cs load into consider. Otherwise, a restart cs may have to clean all it's replica cz they have all been recovered to other chunk servers.
当前冗余副本要么不删，要么见一个删一个
没考虑负载均衡，如果一个机器停了一段时间再起来，那其上的文件都会是4副本，report时会被全删，而其他3个副本所在机器负载一点都不会被分担。

No files, tests, or entry points are named. Trace the flag controlling redundant-replica removal and the replica cleanup performed during recovery and report, then determine how chunk-server load should influence deletion; done means cleanup avoids concentrating load after a restarted server rejoins.

Written by the indexing model from the issue text.