fix: preserve falsy but valid multi-index level names

[Kropiunig]

What happened

Building a pandas.MultiIndex-backed index whose level name is falsy but valid — most naturally an empty string "" — raises KeyError through public APIs:

import numpy as np, xarray as xr

ds = xr.Dataset(
    {"v": (("x", "y"), np.arange(6).reshape(3, 2))},
    coords={"": ("x", [10, 20, 30]), "b": ("y", [1, 2])},
)
ds.stack(z=["", "b"])        # KeyError: 'z_level_0'
# ds.set_index(...) with a "" level name fails the same way

An empty string is a perfectly valid coordinate name (ds.sel({"": 20}) works, "" in ds.coords is True), so this is a crash on legitimate input rather than rejection of a disallowed one.

Cause

PandasMultiIndex.__init__ fills in default names for unnamed levels:

name = idx.name or f"{dim}_level_{i}"

pandas represents an unnamed level as None, but or also discards legitimate falsy names ("", 0, False), swapping them for the synthetic {dim}_level_{i}. Meanwhile level_coords_dtype is still keyed by the original name, so the two fall out of sync and create_variables raises KeyError on the lookup self.level_coords_dtype[name].

Fix

Use an explicit is not None check so only truly unnamed (None) levels get the default name:

name = idx.name if idx.name is not None else f"{dim}_level_{i}"

Tests

Added a regression assertion to TestPandasMultiIndex::test_constructor covering a "" level name (fails before, passes after). The existing default-name case (None levels → x_level_0, x_level_1) is unchanged and still asserted.