Benchmarks wishlist

[TomAugspurger]

What high-level areas do we want coverage in?

• [x] dask.order.order
• [ ] Anything from dask.core?
• [x] dask.optimization?
• [ ] task stealing?
• [ ] scheduler throughput on large graphs?
• [ ] ...

cc @mrocklin @jrbourbeau @eriknw (I think Erik has some benchmarks for dask.order.order)