[Do not merge] Benchmark Tokio executor vs Rayon

[michaelsproul]

This is an experimental PR to try using Tokio in place of Rayon. Tokio also uses a work-stealing thread pool for execution, but has better support for mutexes and locks (see rayon-rs/rayon#592).

It turns out that these better locks come with significantly higher overhead, and tree hashing runs 3-4x slower on Tokio than it does on Rayon: 140ms (tokio) vs 40ms (rayon).

This branch needs to be compiled with tokio_unstable enabled:

RUSTFLAGS="--cfg tokio_unstable" cargo bench -- async_tree_hash_root_list