feat(interactive): Introduce explicit barrier for actors when switching query service to a different graph

[zhanglei1949]

When a StartService(graph) is received by admin service, we will

1. The actor scope for current query actors will be cancelled.
2. During the scope cancellation process of the query actors or after scope cancellation is completed, all requests sent to the query_service will fail and be rejected.
   The response of the http request will be like

{
  "code": 500,
  "message" : "Unable to send message, the target actor has been canceled!"
}

3. After the previous graph is closed and new graph is opened, the new query actors will be available in a new scope.
4. The query service is now ready to serve requests on the new graph.

Fix #3394

[LiSu]

While the process of scope cancellation is ongoing, the messages encapsulating newly arrived query requests (sent to the cancelled actor) will be deleted in the current implementation, is this the expected behavior?

[zhanglei1949]

Yes, we will only process the requests received before the cancellation of actor scope. Requests received during the cancellation and after the cancellation is expected to return immediately.

{
  "code": 500,
  "message" : "Unable to send message, the target actor has been canceled!"
}

[LiSu]

LGTM