Feedback on neural classification lab

• Don't label sections "Part 1" etc: the css automatically numbers headings.
• Allow the --explore flag to take a number--how many examples to show.
• Rename models.handpicked to models.features, like in the previous lab.
• Add an option --save, to save the configuration and weights of a model into a directory (weights?). Add this directory to gitignore. Whenever a fully-qualified module name would be used in the CLI to specify a model, the path to a saved model can be used instead.
• Add an option, --run, which turns on the computer's webcam and starts applying the specified classifier in a loop forever, using the webcam's input, printing the predicted class to the terminal, along with confidence. The idea is that once we have a working classifier, students can hold up their own handwritten digits and get them classified. This should be in the last section of the lab.

Also, go through the curriculum page and respond to inline feedback--fixing and removing comments where you can, and responding inline when discussion is needed.