Update releaser to use Argo.

[jlewi]

Our releaser for tf-operator is currently broken.

Due to various issues.

HEAD is now at 330eb92... Add an option to release.py to specify the tag for the image to use. (#357)

INFO|2018-02-23T00:51:58|py/util.py|79| Running: git rev-parse HEAD 
cwd=/tmp/tfk8s.src.tmp.g8aL4E
INFO|2018-02-23T00:51:58|py/util.py|88| Subprocess output:
330eb9239e006c2bcd04f90d57b01f873124c3e9

+ GOPATH=/tmp/tfk8s.src.tmp.g8aL4E/go
+ mkdir -p /tmp/tfk8s.src.tmp.g8aL4E/go
+ cd /tmp/tfk8s.src.tmp.g8aL4E
+ python -m py.release build_new_release --src_dir=/tmp/tfk8s.src.tmp.g8aL4E --registry=gcr.io/tf-on-k8s-dogfood --project=tf-on-k8s-releasing --releases_path=gs://tf-on-k8s-dogfood-releases
INFO|2018-02-23T00:51:59|/tmp/tfk8s.src.tmp.g8aL4E/py/release.py|469| Latest passing postsubmit is 330eb9239e006c2bcd04f90d57b01f873124c3e9
INFO|2018-02-23T00:52:00|/tmp/tfk8s.src.tmp.g8aL4E/py/release.py|472| Most recent release was for cabc1c0-dirty-e3b0c44
INFO|2018-02-23T00:52:00|/tmp/tfk8s.src.tmp.g8aL4E/py/release.py|330| Use --src_dir=/tmp/tfk8s.src.tmp.g8aL4E
INFO|2018-02-23T00:52:00|/tmp/tfk8s.src.tmp.g8aL4E/py/release.py|339| /tmp/tfk8s.src.tmp.g8aL4E/go/src/github.com/tensorflow/k8s does not exist.
INFO|2018-02-23T00:52:00|/tmp/tfk8s.src.tmp.g8aL4E/py/release.py|346| Creating symbolic link /tmp/tfk8s.src.tmp.g8aL4E/go/src/github.com/tensorflow/k8s pointing to /tmp/tfk8s.src.tmp.g8aL4E
INFO|2018-02-23T00:52:00|/tmp/tfk8s.src.tmp.g8aL4E/py/release.py|369| vendor directory exists; not installing go dependencies.
Traceback (most recent call last):
  File "/usr/lib/python2.7/runpy.py", line 162, in _run_module_as_main
    "__main__", fname, loader, pkg_name)
  File "/usr/lib/python2.7/runpy.py", line 72, in _run_code
    exec code in run_globals
  File "/tmp/tfk8s.src.tmp.g8aL4E/py/release.py", line 700, in <module>
    main()
  File "/tmp/tfk8s.src.tmp.g8aL4E/py/release.py", line 697, in main
    args.func(args)
  File "/tmp/tfk8s.src.tmp.g8aL4E/py/release.py", line 480, in build_new_release
    build(args)
  File "/tmp/tfk8s.src.tmp.g8aL4E/py/release.py", line 373, in build
    build_and_push(go_dir, args.src_dir, args)
  File "/tmp/tfk8s.src.tmp.g8aL4E/py/release.py", line 393, in build_and_push
    version_tag=args.version_tag)
AttributeError: 'Namespace' object has no attribute 'version_tag'

I think we should refactor this to be an Argo workflow that builds ontop of our E2E workflows so that we validate the image as part of a release.

We could start with what our E2E tests do and then if the tests pass push the image to an appropriate registry with the appropriate tag.

[jlewi]

I don't think it makes sense to update the cron job/releaser that regularly rebuilds the TFJob operator image. We should just delete that code right now since its no longer used.

Instead of updating that, we should think about how to continuosly buiild docker image images as needed as part of our release process for Kubeflow ksonnet packages.

Per kubeflow/testing#40 right now we would just manually run the Argo workflow to build the TFJob image before cutting the ksonnet release.

[gaocegege]

SGTM, I think we should establish the continuous deployment process to build and publish docker image.