Scalar non-dimension coords forget their heritage

I'm not sure if this is a bug or a feature, to be honest. I noted that expanded scalar coordinates would retain their scalar values, and so thought that non-dimension coordinates might keep their references to each other and do the same.

What happened:

When a dimension is squeezed or selected with a scalar, the associated non-dimension coordinates are unassociated from the dimension. When the dimension is expanded later, the previously associated non-dimension coordinates are not expanded.

What you expected to happen:

The previously associated non-dimension coordinates should be expanded.

Minimal Complete Verifiable Example:

import numpy as np
import xarray as xr

arr1 = xr.DataArray(np.zeros((1,5)), dims=['y', 'x'], coords={'e': ('y', [10])})
arr2 = arr1.squeeze('y').expand_dims('y')

xr.testing.assert_identical(arr1, arr2)

Error:

AssertionError: Left and right DataArray objects are not identical

Differing coordinates:
L   e        (y) int64 10
R   e        int64 10

Taken another way, I would desire these statements to be possible:

xr.DataArray(np.zeros(5), dims=['x'], coords={'y': 0, 'a': ('y', 1)})
xr.DataArray(np.zeros((0, 5)), dims=['y', 'x'], coords={'e': ('y', 10)})

Environment:

Start by running the minimal example and inspecting the behavior of DataArray.squeeze('y') followed by expand_dims('y'). Trace how the non-dimension coordinate e is handled when its dimension is removed and restored. Done means the round trip passes xr.testing.assert_identical and the expected scalar-coordinate cases remain valid.