Always use DropoutScheduler

We always use DropoutScheduler, we should remove the parameter from the training configuration. Ref https://gitlab.teklia.com/atr/dan/-/blob/1956acc9bbe79f6ce1b8ffba69fd3fe16cb49571/dan/manager/training.py#L189

changed milestone to %DAN-P5: Improve training

added P2 label

added Quick Win label

assigned to @yschneider

mentioned in merge request !233 (merged)

closed with merge request !233 (merged)