Run a pipeline without dev_mode (defaults to False); You get a dataset named {dataset_name}

Repro

1. Set dev_mode=True. For each run, it produces a namespaced dataset {dataset_name}_{timestamp}.
2. Set dev_mode=False, data is written to the latest namespaced dataset.

Intuitively, Step 2 should create or write to non-namespaced dataset named {dataset_name}.

I don't see good a reason why the sequence dev_mode=True -> dev_mode=False should matter. The behavior of dev_mode=False should be consistent (not stateful)

I'm filing this as a bug given behavior isn't intuitive.