Fix and refactor color handling in DecisionBoundaryDisplay #33115
Description
This is to keep track and address several things we noticed in #33015
I will address them in follow-up PRs.
- check number of (distinct)
colors == n_classes - use same selection mechanism for all qualitative cmaps (that have
colors) - refactor and unify checks (some done in
from_estimator, some inplot) - use colors for binary problems as well (but keep
viridisas default) - maybe even move color selection into a separate function?
Please add, if I missed anything @lucyleeow @ogrisel