Change the way how the crash loop buggify setting works

[johscheuer]

What would you like to be added/changed?

In the current implementation we change the arguments to crash-loop, this has the drawback that the Pod has to be recreated. In the recreation step (deletion and creation) the cluster state could be modified, so that the Pod is pending for some time. In addition, the recreation can take some time. A better and faster solution would be to change the image to a non existing image like localhost/foundationdb/crash-loop, this will keep the Pod running and will result in a similar setup.