Rebuild/bump kubeadm CI image.

[pipejakob]

I wanted to rebuild this image to take advantage of some recent merges (#2094, #2179, #2182, #2183).

Based on my local testing, this should bring the kubeadm e2e job back to green.

[krzyzacy]

/lgtm

[krzyzacy]

FYI seems the job is more broken now :-(

[pipejakob]

So what I'm seeing is that when I converted this job from Jenkins to prow (a while ago), I kept the same name, but prow restarted the job run numbering, so new runs of this job are actually showing old logs from Jenkins runs back in January, since it's reusing old job run numbers. Here's an example of a fresh run that actually has logs from back in January, so they're not actually valid:

https://k8s-gubernator.appspot.com/build/kubernetes-jenkins/logs/ci-kubernetes-e2e-kubeadm-gce/464

I've used gsutil to delete some old logs around the upcoming sequence numbers to see if that would let new runs upload their logs, but now I'm getting no logs for those runs:

https://k8s-gubernator.appspot.com/build/kubernetes-jenkins/logs/ci-kubernetes-e2e-kubeadm-gce/467

I need to dig into why new runs are completely missing logs.

[pipejakob]

To close the loop on this, I was able to capture logs from a live run of the job and found that GOOGLE_APPLICATION_CREDENTIALS wasn't being set. I used kubectl describe pod and verified that it wasn't a typo, that some of the environment variables were just missing. I have a PR for the fix now: #2246.