GitLab QA For Auto DevOps Improvements

Below are a list of follow up tasks that will make the tests better for Auto Devops. These are a follow up from #112 (closed) so that we can keep the first iteration simple.

• Improve cleanup steps in script. Apparently we need to delete load balancers separately from cluster as well as persistent disks as we pay for both of these. (need to figure out how to do this...) See #297 (comment 87272971) for instructions on how to delete orphaned load balancers
• Improve our ability to debug when a particular build step fails. (eg. #289 (closed) ) . One possible option is to do job passing assertions on the job page rather than pipeline page so that the log output is included in the screenshots. Another option would be to implement more logic in the assertion code so that if the job status is wrong it will switch to that page and take screenshots of that page as well as pipeline page (assuming pipeline page view is useful)
• GDK setup and instructions. We could generate a unique URL during gdk reconfigure, add this URL to gitlab.yml as hostname and add the SSH tunnel to Procfile so that you have an internet accessible route to your application just by doing echo true > auto_devops && gdk reconfigure
• Try to eliminate the risk we create a K8s cluster in a project we shouldn't (since we are using local gcloud CLI if people run bin/qa)
• Add Assertion To Auto DevOps Spec About Running Production Application
• Test review applications
• Test running canary deployments
• Test the performance metrics page shows properly
• Perhaps use google cloud API instead of gcloud CLI as this would allow better error handling
• Install cluster applications in parallel. I tried this before but it didn't work for some reason so I put this off until later but it should be done since it will save minutes of waiting probably.
• Schedule a cleanup job on Kubernetes to delete the cluster in case it isn't deleted in the after block
• Add a test to confirm that JupyterHub installs properly