The ShardRegistry periodically schedules an internal reload on its own dedicated executor. However, this reload is not interruptible in any way, even though shut down of the ShardRegistry will join on it.

Given that this internal reload has no operation deadline, it can get stuck for various reasons, which in turn will block the shutdown sequence.

As part of this ticket, we should make ShardRegistry keep track of its internal reloads (possibly through an OperationContextGroup) so that it can be interrupted on shutdown.