* Start expanding runner.py to run an actual E2E test.

…s/master/test/helm-test/main.go.

This will serve as the basis for deploying and running the tests.

* Deploys the TfJob helm chart and then runs the tests.
* Doesn't deploy the chart in its own namespace because the E2E test
  assumes its using default namespace.

* The output can be used by runner.py to determine the GCR path of
  the newly built image.

  and run the tests on.

* Fix a lint issue in the helm chart.

* Need to install docker so we can build the image locally without GCB.
* Fix some flags.
* runner.py should delete the GKE cluster.
* Add instructions about running the test inside a Docker container.

* Need to get credentials otherwise helm won't work.
* Echo command output to logs so that it will show up in job log.
* Add some notes on the steps followed to setup Prow integration.

* Run helm init as part of retry loop waiting for helm to be initialized.
* Set a 2 minute timeout on waiting for helm to be initialized.

up running in namespace test-pods.

* Need to upgrade six to support google.cloud.storage library

Tests should pull runner.py from the repository and not the image.

so we know which image a job ran in.