ENH: kurtosis adjustment for variance tests, Bartlett

[josef-pkt]

just a reminder issue (I don't have the references right now)

Bartlett variance test can be improved for non-normal distributions by taking kurtosis into account.

One of the monte carlo articles for #6526 included the kurtosis corrected Bartlett test

There is also a kurtosis correction in multivariate methods that I looked at, another issue (?)