Hi,

With the release of zarr-python 3, the feature to store multiple chunks in a single storage object (i.e. sharding) was introduced. Since, Dask has started to support zarr-python 3, I would request the option to specify the the shards in the to_zarr() function be introduced. This will be a very helpful feature for people trying to store large arrays on regular file systems while needing fast random reads.