[RayService] No serve deployments after head node failure

[simon-mo]

The re-deployment did not happen when the head node went down and came back again. I think we should do a re-deployment regardless of whether GCS HA is enabled or not (to reconcile the state after all)

Steps to repro:

• apply the sample app
• kubectl delete head pod
• observe head pod came back
• observe head pod has serve controllers and http proxy
• observe head pod doesn't have any other serve actors
• the controller log show that shouldUpdate is false

might be related to #374