I would second the use of datatree. The GEDI data structure should be quite similar to ICESat-2 (since they’re both laser altimeter instruments). Check out this comment at Reader.load() issues with ATL06 - #2 by weiji14 on using datatree.open_datatree.
Oh, and if you’ve got some working code on getting GEDI data from HDF5 to Zarr, I would love to check it out! There’s some discussion from the NASA Science team on getting a cloud-optimized (e.g. Zarr or other) format for ICESat-2 (see GitHub - CryoInTheCloud/IS2CloudOptimizedData: For work to build a cloud-optimized data format standard for NASA ICESat-2 data, nothing much there yet though), and I’m wondering if GEDI might be close enough as a point cloud data structure to get involved in some of the discussion.