Extend preprocessing to deal with multiple dataloader dims

Currently, the preprocessing classes can only deal with a single dataloader dimension, so when you want to scale the inputs x and targets y you would have to fit two different scalers.

This should be able to be automated so that the scaler fits the according values (min, max, mean etc.) separately to all dataloader dims, so that the following syntax should be able:

scaler = MinMaxScaler(data_loader=dataloader)
scaler.fit()

for batch in dataloader:
    x, y = batch
    x_scaled, y_scaled = scaler.transform([x,y])
    preds_scaled = model(x_scaled)
    loss = loss_func(preds_scaled, y_scaled)
    x_rescaled, preds_rescaled = scaler.inverse_transform(x_scaled, preds_scaled)