What

This pull request resolves #254. The
following fixes and improvements were made:

• Fix object deletion upon completion: apparently, the main cause of the
  instability was the use of the record to determine the time at which the Run
  object had been updated. I changed the watcher to use the
  Status.CompletionTime since this approach seems to be more
  reliable. Currently, if the Tekton
  pipelines controller resyncs the object and transitions the status, the object
  will change, the record will be updated and this will cause discrepancies in the
  verification of the grace period expiration.
• The aforementioned change fixes end-to-end tests. Now, the test suite is
  passing consistently in my machine (this
  pr should be merged as well
  since it fixes other issues in the current e2e setup).
• Improve logs emitted by the watcher: favor structured logging since it
  simplifies a lot the debugging with log indexer tools such as Splunk and others.
• Refactor the controller's code to make it simpler and hopefully, easier to
  reason about. I tried to make things more uniform and meaningful for new
  collaborators. One of the most relevant changes was removing the enqueue
  function that was being passed around and replacing the approach with the
  knative's controller.NewRequeueAfter function which returns an error that
  tells knative to requeue the key after a given duration.