Fix rank biserial correlation sign for Mann-Whitney U test

[thimo72]

Changed RBC calculation from rbc = 1 - (2 * uval) / diff.size to rbc = 2 * cles - 1

This makes RBC directional and consistent with CLES, so that RBC > 0 when x > y.

[raphaelvallat]

Hi,

Thanks for opening the PR. Is this linked to a particular GitHub issue/discussion? If not, can you please first open an issue to describe the problem, with minimal code to reproduce the issue?

Also, for this PR, please add unit tests to validate the new behavior.

Thanks Raphael