Convergence diagnostic for importance sampling

[matsen]

I'm just going to use 500 samples, which looks pretty OK:

But it would be good to be able to check for convergence.