Update Learners.ipynb to the current deep-learning API (#1257)

The notebook called the old functional API (perceptron_learner,
neural_net_learner, gradient_descent, learning_rate=), which no longer
exists since the deep_learning4e refactor to classes, so Cell 4 raised
NameError: name 'perceptron_learner' is not defined. Port the cells to
PerceptronLearner/NeuralNetworkLearner(...).fit(X, y) with
stochastic_gradient_descent and l_rate=, building X/y from the datasets
as the tests do, and pass an explicit hidden-layer list to the neural
network. Also define the illustrative input_size/output_size/
hidden_layer_sizes used by the raw_net snippets. Notebook now runs
top-to-bottom with 0 errors.

Author: dmeoli