Add back service informer to handle Service recreations quicker

[jessesuen]

A user did the following:

1. Created a blue/green Rollout with active and preview services.
2. Deploy the services and Rollout. Wait for Rollout to fully deploy
3. Delete the services
4. Redeploy the services. The services will be missing the pod hash label added by Rollout

After step 4, because we no longer have Service informer, we do not detect that the services are missing the pod hash. As a consequence, both the active and preview services will be servicing traffic from both the active and preview deployments.

A while back we made a conscious decision not to add a Service informer because we felt user would not be deleting the pod-template-hash labels, and it would save a API server watch on services. However, I did not consider the service delete/recreate use case, which is what happened here, and may be common/desirable to do.

In order quickly correct situations where a Service was deleted and recreated (or pod template hash labels were removed), we should consider adding back the Service informer to immediately restore the labels.

The tricky part is knowing how efficiently associate Service updates which are related to Rollouts (since they would not have any our labels yet).

[alexmt]

Working on it