Baseline shell benchmarks
With trace improvements recently merged, we've been looking at CPU and memory profiles.
Now that trace performance is no longer an issue, it's probably a good idea to add baseline benchmarks for other hot code paths that have the potential to become issues or we believe we can further improve.
One hot code path is the shell implementations. Every single job requires the script for various stages to be generated.
We should add benchmarks that cover each shell and scenarios that cover the most common code paths.