When using delegates to run Harness jobs, they all use the root volume of the node for their ephemeral disk space. This causes issues with I/O when many jobs are placed on the same node. We would like to be able to specify that jobs use ephemeral PVCs instead, but the pipeline configuration does not support this