Arvados

Individual CWL steps don't have a cwl.input.json file to display. However, we can have arvados-cwl-runner to put cwl input contents into the same cwl.input.json mount point that is used for workflows. However, unlike workflows, changing the contents of cwl.input.json and re-running the step would not have any effect, because the parameters have already been baked into the command line and mount points.

Somewhat related, we should have a convention for identifying intermediate output collections, so that they can be more easily filtered, selected, and trashed.