Snapshot based dataset splitters should delete definitions or add datestamps(uuids?)
We have had some failures of dataset splitters because they are trying
to redefine datasets with the same name but different definition text
(usually with different snapshot ids, etc.) I think this particularly
happens with multiple campaigns splitting the same dataset...
They should either delete and recreate them, or add more uniquifying info
to the dataset name, so we get distinct definitions. Probably a timestamp
is sufficient, but we could go all UUID on it...