Train with Generator

[JackyP]

Referencing discussion in https://github.com/mlr-org/mlr3keras/issues/6.

• This requires splitting the recently added x_transform and y_transform further into getting the data from the task and the actual transformation itself,
• class_weights should be class_weight according to latest keras documentation,
• Train and valid splits for the generator using mlr3 resampling.

[JackyP]

Let me look into how it fits in with latest changes in tabnet branch...

[JackyP]

Superseded with https://github.com/mlr-org/mlr3keras/pull/12