We could add some APIs to allow access to the files in a BlobTree without the user having to actually download the whole dataset. That metadata is available.

Just a sketch, but:

1. JuliaHub.dataset_filetree(dataset) -> FileTree: returns a data structure that can be used to iterate over the file tree (maybe based on https://github.com/shashi/FileTrees.jl).
2. JuliaHub.download_dataset_file(dataset, filepath, dstfile): downloads an individual file from a BlobTree, without having to download the whole BlobTree in one go.