Rule `no-empty-chunks` is not yet useful

[forman]

Describe the bug

The rule no-empty-chunks ...

1. outputs a wrong message. Instead of 'write_empty_chunks=True' it should read 'write_empty_chunks=False'.
2. should not be applied to coordinate variables by default because coordinates are often not chunked.
3. should be applied to chunked variables only. It seems, it emits also for single chunk variables.

valid.zarr:
dataset.coords['x']               warn  Consider writing the dataset using 'write_empty_chunks=True'.  no-empty-chunks
dataset.coords['y']               warn  Consider writing the dataset using 'write_empty_chunks=True'.  no-empty-chunks
dataset.data_vars['sst']          warn  Consider writing the dataset using 'write_empty_chunks=True'.  no-empty-chunks
dataset.data_vars['sst_anomaly']  warn  Consider writing the dataset using 'write_empty_chunks=True'.  no-empty-chunks
dataset.coords['time']            warn  Missing 'units' attribute in variable 'time'.                  var-units

To Reproduce

Run the notebook notebooks/xrlint-cli.ipynb.

[forman]

In xarray, the write_empty_chunks=False setting affects how the dataset is written to disk, specifically when using to_zarr(). However, once the dataset is written, this setting is not stored as metadata in the dataset itself or in its encodings. Instead, we can infer whether write_empty_chunks=False was used by checking if empty chunks are missing in the underlying Zarr store. But this can be an expensive operation, especially if data is in remote storage.