Support a new deletion policy to stop first

In some user cases, user want to stop the nova server first to do some cleanup work before to terminate the server, one example:
Magnum using heat to deploy kubernate cluster, each node will register to the mater node when creation, and the ones to be terminated should to logout first.
But now in heat, if the count of ResourceGroup is scale-down, the resource is deleted directly, because we don't have the corresponding deletion policy to stop first before terminate. This blueprint aim to add such deletion policy.