MMD demos would be cool

[karlnapf]

@sejdino @Saurabh7 @lambday

Let's think of something nice for that. We could have parameters of a normal/Laplace/KDE distribution that is compared to something else. And then we can plot null distribution of MMD. Like in the notebook. Let me know your thoughts