Don't rely on out_features to get output dimension of PytorchModel

[Robbybp]

out_features is only defined for linear layers (or, at least, it's not defined for sigmoid layers). MWE:

# model.py
import torch
nn = torch.nn.Sequential(torch.nn.Linear(1, 10), torch.nn.Sigmoid())
torch.save(nn, "sigmoid_output.pt")

# model.jl
import MathOptAI
import PythonCall
import JuMP

nn = MathOptAI.PytorchModel("sigmoid_output.pt")
gb = MathOptAI.GrayBox(nn)
model = JuMP.Model()
JuMP.@variable(model, x)
MathOptAI.add_predictor(model, gb, [x])

Fails with: ERROR: LoadError: Python: AttributeError: 'Sigmoid' object has no attribute 'out_features'