Make SECFisher more accurate

[eviatarbach]

Currently, SECFisher uses the sample correlation $r$ as the best estimate for the true correlation. However, due to the skew of the sampling the distribution, this is not the case. Instead, the true correlation $\rho$ can be better approximated as $\rho\approx \frac{\exp(2\text{arctanh}({r})) - 1}{\exp(2\text{arctanh}({r})) + 1}$.

[eviatarbach]

Actually, this is equivalent to plugging in the sample correlation directly.