BUG: numpy.cov doesn't keep dimensions for matrix with exactly one column

Describe the issue:

Expected Behavior:

np.cov(X, rowvar=False) returns array with same number of dimensions as X

Actual Behavior:

np.cov(X, rowvar=False) returns array with zero dimesions if X.shape[1] == 1

Reproduce the code example:

import numpy as np

# keep dimensions as expected
a = np.random.rand(100, 0)
assert np.cov(a, rowvar=False).ndim == a.ndim

c = np.random.rand(100, 2)
assert np.cov(c, rowvar=False).ndim == c.ndim

# doesn't keep dimensions
b = np.random.rand(100, 1)
assert np.cov(b, rowvar=False).ndim == b.ndim

Error message:

No response

NumPy/Python version information:

1.21.5 3.7.3 (default, Jan 22 2021, 20:04:44)
[GCC 8.3.0]

Start at the np.cov entry point and trace the rowvar=False path for an input whose shape is (100, 1). Reproduce the reported assertion, then add coverage for the one-column case and confirm that the result preserves the input dimensionality without regressing the zero- and two-column examples.