Deprecate open_zarr in favor of open_dataset(..., engine='zarr')

[jhamman]

What is your issue?

We have discussed many time deprecating xarray.open_zarr in favor of xarray.open_dataset(..., engine='zarr'). This issue tracks that process and is a place for us to discuss any issues that may arise as a result of the change.

xref: #2812, #7293

[keewis]

To get back to this, I believe we should at least harmonize the meaning of the chunks parameter for both functions: open_zarr currently translates "auto" to {} if a chunk manager is available, or None otherwise, then forwards everything to open_dataset. open_dataset deals with None separately, but passes "auto" and dict values on to dask (with several translation layers in between), which means that open_dataset(..., chunks="auto") is not the same as open_zarr(..., chunks="auto").

I believe we should aim to move the special-casing of "auto" to open_dataset while choosing a string that does not conflict with the chunks parameter for dask's rechunk.

[jhamman]

👍 I have also come around to keeping open_zarr.

Your plan around the chunks parameter seems quite reasonable.