FNN class constructor does not contain regularization implementation for PyTorch backend

[cwoolfo1]

The FFN class constructor for the PyTorch backend does not contain the option to set regularization in its initialization despite having the regularizer as a attribute of the nn base class.

I had to set it manually in my code

net = dde.nn.FNN(layer_size, activation, initializer)
net.regularizer = ["l2", 1e-6]

[lululxvi]

Feel free to submit a PR to improve the code.