Learners should properly handle new offset column role in Task

[bblodfon]

See mlr3 PR for the offset column role implementation.

The following learners that have an offset parameter:

• In mlr3learners :
  • glm => stats::glm() (lrn("classif.log_reg")) - binary class only, stats::lm() (lrn("regr.lm"))
  • {classif|regr}.{cv_}glmnet learners have offset during train and newoffset during predict (see stackoverflow issue) - binary + multiclass
  • xgboost learners (classif.xgboost - binary + multiclass + regr, regr.xgboost)
• In mlr3extralearners:
  • lme4::lmer() - lrn("regr.lmer")
  • mgcv::gam() - lrn("classif.gam") (binary) + lrn("regr.gam")
  • gamboost and glmboost => mboost_* (classif - binary + regr)
  • partykit tree and forest (e.g. partykit::cforest()) learners in mlr3extralearners?
  • lrn("regr.glm")
  • survival glmnet learners
  • survival BART
  • dbarts learners
  • earth learners???