Fix doc of windowed_mean|variance to match implementation

[nicolaspi]

Subsumed by #1600

Hi, I got confused with the results of windowed_variance, it seems the documentation is wrong (same applies for windowed_mean).

low = np.maximum(np.arange(10) - 5, 0)
high = np.arange(10) + 5
x = np.arange(20, dtype=np.float32)
a = tfp.stats.windowed_variance(x, low_indices=low, high_indices=high)
b = [np.var(x[low[i]: high[i]+1]) for i in range(len(low))]
c = [np.var(x[low[i]: high[i]]) for i in range(len(low))]

np.testing.assert_almost_equal(a, c, decimal=5) # Ok
np.testing.assert_almost_equal(a, b, decimal=5) # Not Ok