I see no good reason for only performing the later phases (PhaseClusterShutdown) when downing a node. The tasks for graceful stopping singleton and sharding can be skipped when in downing state.

This makes it easier to perform tasks for earlier phases (e.g. service unbind) also when downing.