[Improvement] Restart cluster components with order defined in ClusterDefinition

[nayutah]

Is your improvement request related to a problem? Please describe.
When restart a cluster, restart the component one by one and following the order defined in ClusterDefinition, this can avoid many concurrency problems. For clusters with more than one interdependent components, like sentinel redis and orchestrator mysql, it is a practical way to make the cluster more robust.
For reconfigure/vscale operation that may incur restarts, it should also follow the defined component orders.

[nayutah]

also helps to fix #7519