registered datasets for stability compute

facilitate research by publishing dataloading wrappers to our hosted datasets so people don't need to keep redownloading and wasting time designing dataloaders to use data we've already figured out how to access efficiently