Set up benchmarks to compare performance of runner backends in different scenarios

[ben-denham]

E.g.

• Lots of small tasks
• Large memory size of results that are dependencies
• I/O bound tasks