- The varnish controller deleted a service from its
map under the wrong conditions.

- Not an error if there are no Endpoints for the
Service when the Delete event is synced -- they
may already be gone.

- Varnish and haproxy controllers delete the service
from their maps before any other sync actions are
taken -- otherwise error returns may prevent the
deletion from ever happening.

- Permanent network errors on attempts to communicate
with the admin interfaces are ignored (do not cause
re-queue), since the instance may be gone.

Starting with a pristine cluster, it initially deploys the controller
and unready Services for Varnish Ingress, and verifies the deployment.

Then each of the examples are deployed, verified, and undeployed.

On exit the Services for Ingress and controller are undeployed,
restoring the pristine state of the cluster.

Deployment is now in the API group apps/v1.

RBAC for the namespace example was missing rules to access resources
in the ingress.varnish-cache.org APIs.