groupby multiple variables should include observed groups only

What happened?

Grouping more more than one variable includes combinations with no observations

What did you expect to happen?

Empty groups are not included

Minimal Complete Verifiable Example

# /// script
# requires-python = ">=3.11"
# dependencies = [
#   "xarray[complete]@git+https://github.com/pydata/xarray.git@main",
# ]
# ///
#
# This script automatically imports the development branch of xarray to check for issues.
# Please delete this header if you have _not_ tested this script with `uv run`!

import numpy as np
import xarray as xr
xr.show_versions()

coords = {'y': [0, 1, 2, 3], 'y1': ('y', [1, 1, 2, 2]), 'y2': ('y', [1, 1, 2, 2])}
xdata = xr.DataArray(data=np.ones((4, 2)), coords=coords, dims=['y', 'x'])

xdata.groupby(['y1', 'y2']).mean()

MVCE confirmation

• Minimal example — the example is as focused as reasonably possible to demonstrate the underlying issue in xarray.
• Complete example — the example is self-contained, including all data and the text of any traceback.
• Verifiable example — the example copy & pastes into an IPython prompt or Binder notebook, returning the result.
• New issue — a search of GitHub Issues suggests this is not a duplicate.
• Recent environment — the issue occurs with the latest version of xarray and its dependencies.

Relevant log output

<xarray.DataArray (y1: 2, y2: 2, x: 2)> Size: 64B
array([[[ 1.,  1.],
        [nan, nan]],
       [[nan, nan],
        [ 1.,  1.]]])
Coordinates:
  * y1       (y1) int64 16B 1 2
  * y2       (y2) int64 16B 1 2
Dimensions without coordinates: x

Environment

Start with the provided Python MVCE and inspect xarray's multi-variable groupby implementation and its tests, using the reported output as the baseline. Verify that grouping by y1 and y2 returns only observed combinations (1,1) and (2,2), without empty combinations such as (1,2) or (2,1).