Introduce MLP and associated variables in XOR-MLP example

As a reminder, x_data and y_data make up the data, the MLP weights and biases are the model parameters, and the locations and scales of the variational normal components are the variational hyper-parameters. There are no model hyper-parameters.