Implement Criterion benchmarks for ext

[JoeyBF]

We already have some hand-rolled and iai-based benchmarks, but I think having a more robust framework would make it much easier to measure improvements in the future, especially if the improvements are small but statistically significant.

@mcy I think you said that was something you would be interested in?