Follow-on to #32743 (in the theme of #30315).

When a shard region gracefully leaves, a rebalance is triggered which results in RebalanceDone messages. Each RebalanceDone, if the shard is still in state, results in a ShardHomeDeallocated event. In the ddata coordinator, while waiting for confirmation most incoming messages are stashed, including Terminated messages.

The shard region leaving will eventually result in a Terminated for that region, which results in a ShardRegionTerminated event, which updates the state as if every remaining shard with that region as home was deallocated (those shards will have ShardHomeDeallocated in their future: the RebalanceDone processing accounts for the ShardRegionTerminated being processed before RebalanceDone.

Accordingly, if there are enough RebalanceDones in a graceful exit (note that if typed cluster sharding is in use, there are likely 1000/cluster_size RebalanceDones: more than 50 is reasonbly likely), the coordinator will spend most of the few hundred milliseconds afterwards waiting for the confirmation from ddata: the current stashing behavior likely means that Terminated is stuck behind RebalanceDones which at this point are superfluous (in terms of updating ddata) in light of the Terminated.

This change prioritizes the first Terminated of a region with shards received while waiting for a ddata update over anything unstashed after that update. Future steps could batch the Terminateds and RebalanceDones, but much of the benefit can be had from This One Trick.