Add feedback weight to `ESNCell`

[MartinuzziFrancesco]

The original formulation for the ESN also involves a feedback connection matrix

$$\mathbf{x}(n+1) = f\!\left( \mathbf{W}\,\mathbf{x}(n) + \mathbf{W}_{\mathrm{in}} \mathbf{u}(n+1) + \mathbf{W}_{\mathrm{fb}} \mathbf{y}(n) \right),$$

for teacher training which is currently not available in ReservoirComputing.jl. Adding this involves adding the feedback matrix Wfb, but also maybe adding a dispatch on train and ESN-family models. It could be like

function train!(rc, (train_data, teacher_data), target_data, ps, st,
        train_method = StandardRidge(0.0);

where train_data are the u s and teacher_data the y s.