Refactor the deployment task (triggering deployment part)

The following discussion from lb-nightly-utils!22 (merged) should be addressed:

• @clemenci started a discussion: (+1 comment)

  I understand what is happening here, but I found it cryptic...

  Looking again at the code I started asking myself why we have

  class DeployBinaries(luigi.Task):
      # ...
      def run(self):
          # ...
          install_id = trigger_deployment(self.project, self.platform)
          # ...
          self.output().wait_for_deployment()

  and not something like

  class DeployBinaries(luigi.Task):
      # ...
      def run(self):
          # ...
          install_id = self.output().trigger_deployment()
          # ...
          self.output().wait_for_deployment()

  or a variation where we use self.input() so that the artifact URL to deploy becomes self.input().uri (which already does the clean up internally).

  I think we can merge this as is and refactor the deployment task another day. What do you think?