To run the entire microbenchmarks corpus in a reasonable time we have to apply settings across the board that limit runtime, overriding defaults in ways that might make the data less useful.

- Many benchmarks can make do with much shorter runtime than today, and still produce stable results
- Some benchmarks really do need more runtime

One size does not fit all, and any default, implicit setting will be a poor fit for most.

Goal is to keep data quality high while getting runtime down enough that we can run the corpus more often on a reasonably sized set of machines.