Performance benchmarks for PRs?

[jgabry]

Should we set something up so that we can detect performance regressions introduced by PRs? For example, suppose that one of the PRs to switch to using functions from posterior (e.g. #290, #305) actually caused running loo() to be much slower because the version in posterior was less efficient (just using as an example, I don't have evidence this is true). Currently we have no way of detecting that when reviewing the PR.

[VisruthSK]

I think this is a good idea. posterior has touchstone benchmarks which we could copy.

[jgabry]

I've actually never looked at those, although I'm aware of them.