Use deployments instead of pods

Pods instead of deployments are, at least, used for:

• Kafka client
• Zookeeper client

Defining them as pod has the disadvantage that they do not restart after shutting down the cluster.

added modul/infrastructure label

Maybe it's also possible to define an on failure behavior such that also the pod will be restarted. A pod would have the advantage that it can simply be accessed by a name, which is much more convenient that as pod created from a deployment.

mentioned in merge request !80 (merged)