Efficient storage for UUID primary keys

**Is your feature request related to a problem? Please describe.** When importing datasets that use UUID primary keys the current default storage mechanism generally stores one feature per folder upto around 16M records, **Describe the solution you'd like** When importing a dataset I would like to choose the path storage structure, for instance I know that my dataset will have a UUID primary key, with less than 1M records, so one level of 256 hex folders would be more than sufficient. **Describe alternatives you've considered** I cant easily convert my dataset from uuid primary keys, **Additional context** I have written a script to re-index my datasets from base64 to hex encoded folders and it greatly reduces the number of git objects for my datasets.