Fix duplicate X-values in KDE distribution

[michael-nml]

This PR addresses an issue for KDE distribution calculation returning duplicate X-values. This is an edge case of the statsmodel.nonparametric.KDEUnivariate function.

Reproduction example:

from statsmodels import api as sm
from matplotlib import pyplot as plt

data = [-0.100865, -0.100865, -0.100865, -0.100865, -0.100865]
kde = sm.nonparametric.KDEUnivariate(data)
kde.fit()

print("nr duplicate X-values:", len(kde.support) - len(np.unique(kde.support)))
plt.plot(kde.support, kde.density)

Output:
nr duplicate X-values: 505

Normally input data containing a single value would result in an exception, but in this case it does not due to some form of precision error.

Running the above example with different input data yields:

• data = np.array([-0.100865, -0.100865, -0.100865, -0.100865, -0.100865], dtype='float16') => error
• data = np.array([-0.100865, -0.100865, -0.100865, -0.100865, -0.100865], dtype='float32') => error
• data = np.array([-0.100865, -0.100865, -0.100865, -0.100865, -0.100865], dtype='float64') => same output as example above

[nikml]

LGTM