More clearly document `symmetrize` for residuals

[nmdefries]

In https://cmu-delphi.github.io/epipredict/articles/epipredict.html, it would be helpful if we can say a little more about how to get nonparametric prediction intervals. For example, do we have to symmetrize residuals? Or not?

Parallels cmu-delphi/epiprocess#368