Add representative benchmarks

[NthTensor]

We need benchmarks to measure proposed improvements against. We probably need several different test cases, each focusing on a different type of workload.

• Fork-join: A compute heavy problem which presents an optimal case for divide and conquer.
• Web-server: A large number of IO-bound futures and asynchronous closures, perhaps simulating a HTTP server.
• Bevy: A large number of compute-bound synchronous closures added to the injector at intervals with a Zipf distribution over duration.

Most workloads will probably be a mix of these three types of tasks, but it makes sense to benchmark them separately in a standard way. When a given benchmark is applicable to rayon as well (eg, no async jobs) then we should make it possible to run with either and compare the two.