Creating subsets (or datasets) with 100,000s accession refs is a performance issue for the backend, especially because the UI request will time out before the server is done processing. The user also cannot know that the server is processing the request and will re-upload the 100,000 refs, using more server resources.
Update the processing of AccessionRef uploads so that they require the current
version of the dataset (or subset) to be correct before adding refs to it.
The UI ticket is genesys-ui#256