Description of problem:

Current implementation of throttling, based on Automate, and should be done in backend, because:

 - Automate sandbox execution of each step and so is expensive
 - Limited access to Automate model attributes
 - Current throttling mechanism is very clumsy
    - Have to ensure its singleton instance
    - Have to manage its life-cycle
    - Have to invoke the throttling logic in every state machine transitions and onEntry/onExit method
    - And only have chance to run when state changes
 - Automate isn’t intuitive in describing migration flow
    - The logic is littered across a bunch of various yaml and ruby files
    - Hard to understand the logic
    - Hard to maintain and enhance