Save z-transformed correlation matrices instead of correlation coefficients

[tsalo]

Summary

This is a request from a user (see this NeuroStars post). This is a straightforward transform, so I don't see a reason to write out both r and z values, but z values are probably more useful to users, so it seems like a worthwhile change.

[tsalo]

xDF can produce the variance of the z-transformed correlation coefficients (Eq. 10 in the paper), so this won't conflict with variance estimation.