BUG/ENH: fix scale.Huber and add robust MScale

[josef-pkt]

see
#4749 bug in Huber
#9171 allow MScale in RLM scale_est, add generic MScale class

This does not include fixes for naming and default inconsistencies in current code.

This now also allows fixed user provided scale in RLM, start_scale
I did not change behavior of init. It's broken unless user sets model.scale before fit.
issue #9211

Also it moves scale_tau from robust.covariance to robust.scale (followup to #8129

[josef-pkt]

that's about finished to keep PR small

maybe more unit tests

aside:
scale_tau was written for ogk and hardcodes bisquare mean and winsorized (huber) scale
I have not looked at generic versions of tau scale for arbitrary psi and chi/rho.
But the same tau scale function is used in regression setting, e.g. DetR package uses it for MM-estimator instead of M-scale.

[josef-pkt]

all green, except unrelated failures

merging