List of major tasks

We should support writing and reading to other distributed storage other than s3
- What storage should we support?
  - local file system
  - gcloud
Dataset reader and metadata should be able to handle empty zarr indices
Ordering of features needs to be set in stone somewhere
- Maybe dont even allow user to pass in features, maybe just the calculated ones.
Weighted sampling should act on a computed array instead of a dask array
extract_patch_extents_and_metadata_in_series should be parallizable now

List of little fixes