issue_comments: 588178645

I did get one test failure TestDask.test_preprocess_mfdataset, which failed with

``` def test_preprocess_mfdataset(self): original = Dataset({"foo": ("x", np.random.randn(10))}) with create_tmp_file() as tmp: original.to_netcdf(tmp)

        def preprocess(ds):
            return ds.assign_coords(z=0)

        expected = preprocess(original)
        with open_mfdataset(
            tmp, preprocess=preprocess, combine="by_coords"
        ) as actual:

          assert_identical(expected, actual)

E AssertionError: Left and right Dataset objects are not identical E Differing dimensions: E (x: 10) != (x: 10, z: 1) E Differing coordinates: E L z int64 0 E R * z (z) int64 0 E Differing data variables: E L foo (x) float64 0.07341 -0.01756 0.6903 ... -0.08741 0.08472 -0.1905 E R foo (z, x) float64 dask.array<chunksize=(1, 10), meta=np.ndarray>

xarray/tests/test_backends.py:2955: AssertionError ```

I think the problem here is that it's happily promoting the coord to a coordinate dimension even though there is only 1 dataset. We should probably only expand dims which are going to be used in the concatenation...