Fix standard deviation of correlated noise

[david-zwicker]

The call to make_correlated_noise should return a field with unit variance, i.e.,

f = make_correlated_noise((10_000,), correlation="cosine")
f().std()

should return a number close to 1, but this does not seem to work. This currently works for all correlations, but cosine.

[david-zwicker]

The problem seems to be connected to the fact that the cosine correlation function is not monotonic, whereas all other correlations are. This might be connected to the Wiener–Khinchin theorem, which seems to imply that the power spectral density should be a monotone function.