Attached is a minimal example, using H21.0.466, for the community that I would have found helpful myself.

Requirement: Need to "group" work items together for processing together, without partitioning.

Reasoning: Avoiding needing to wait until all upstream work items are complete to "group" them together. Depending on the situation, not using partitioning could be advantageous if you don't want to bottleneck your graph.

Note: I'm muddling through this stuff. There probably is a better way to pull this off. If you know if a better way, please post it up and add to the conversation.