pymc.dims does not impute missing (NaN) observed data

Unlike pm.Normal, a pymc.dims distribution given an observed DataArray containing NaNs builds the model but returns a non-finite logp instead of creating an imputation RV for the missing entries.

import numpy as np
import xarray as xr
import pymc as pm
import pymc.dims as pmd

Y = xr.DataArray([[1.0, 2.0], [np.nan, 3.0]], dims=("a", "b"))
with pm.Model(coords={"a": [0, 1], "b": [0, 1]}) as m:
    a = pmd.Normal("a", 0, 1, dims="a")
    pmd.Normal("obs", a, 1.0, observed=Y, dims=("a", "b"))

m.compile_logp()(m.initial_point())  # nan

Start with the pymc.dims Normal distribution path when an observed DataArray contains NaNs, using the provided model and pmd.Normal calls as the reproducer. Confirm that missing entries create an imputation RV and that m.compile_logp()(m.initial_point()) returns a finite value for the example.