storage: Priority for load-based rebalancing

[bdarnell]

Replica snapshots currently use two priority buckets: recovery (for ranges below their target for live nodes) and rebalance (for all other moves). These are rate-limited to 8MB/s and 2MB/s respectively. Load-based rebalancing should be prioritized higher than count-based rebalancing, and probably use the higher "recovery" rate limit. (within the bucket we can further prioritize based on the degree of imbalance)

[bdarnell]

The distinction between these two buckets has mostly been eliminated: they still use two separate cluster settings, but the defaults for both are the same, and in general we recommend that when users change one they change both. (and the introduction of learner replicas means that the "rebalancing" bucket is now used much more often than the "recovery" bucket). This kind of coarse-grained prioritization no longer seems desirable.