Performance benchmarking

[jdidion]

Compare to rayon, workerpool, and others?

[jdidion]

Initial benchmarks are added in v0.2.1.

It looks like, at least when there are many short-lived tasks, runtime actually increases with the number of threads. This is probably due to thread contention, most likely when locking the task channel receiver to get the next task.

[jdidion]

With longer-running tasks (e.g., one the does thread::sleep(Duration::from_millis(1))), runtime seems to reduce roughly linearly with the number of threads.