Fix complex sample variance

This fix is necessary to make samples consistent with the Hamiltonian. Fixing the variance this way makes np.std(s) of a samples s with variance 1 havyield sqrt(2) and not 1. However, this fix is more compatible with the definition of the complex Gaussian according to wikipedia, and compatible with the definition of the Fisher metric.

nifty6/random.py

@@ -238,8 +238,8 @@ class Random(object):
             raise TypeError("mean must not be complex for a real result field")
         if np.issubdtype(dtype, np.complexfloating):
             x = np.empty(shape, dtype=dtype)
-            x.real = _rng[-1].normal(mean.real, std*np.sqrt(0.5), shape)
-            x.imag = _rng[-1].normal(mean.imag, std*np.sqrt(0.5), shape)
+            x.real = _rng[-1].normal(mean.real, std, shape)
+            x.imag = _rng[-1].normal(mean.imag, std, shape)
         else:
             x = _rng[-1].normal(mean, std, shape).astype(dtype, copy=False)
         return x