On the nature and use of models in network neuroscience