`train!` should not have a bang

[MartinuzziFrancesco]

It doesn't modify anything in place

[MartinuzziFrancesco]

I also think that train should be restructured a bit for v1.0

train(esn, input_data, target_data, ps, st;
    objective = StandardRidge(0.0), # <- what are we using as loss
    solver = QR() # <- what are we using to obtain the solution
) 

This means that also StandardRidge will be modified, although that change is not breaking. Ideally the final version is compatible with both MLJLinearModels and LinearSolve