GETs the root zarr.json and parses its inline
consolidated_metadata.metadata as the node index, so it targets the
Zarr v3 layout that dafr writes. A legacy v2 store (root .zmetadata,
no zarr.json) is rejected with a python -m zarr v2_to_v3 conversion
hint; a v3 store that lacks inline consolidated metadata cannot be
enumerated over HTTP and is likewise rejected. See HttpStore for the
supported alternatives.