Describe the workflow you want to enable

I would like to specify a baseline estimator like in GradientBoostingRegressor(init=MyCoolBaselineEstimator_Maybe_a_linear_model).

Describe your proposed solution

Add a parameter baseline (or if has to be the same, then init) to HistGradientBoostingRegressor and HistGradientBoostingClassifier.

Describe alternatives you've considered, if relevant

The StackingClassifier does not do the same thing.

Additional context

In particular a linear model as baseline could

1. Improve fit time
2. Allow for non-constant extrapolation on important features, quite in contrast to trees.