Benchmark code for the runtime

[toddaaro]

We need a solid suite of simple benchmarks so we can evaluate the performance of the runtime. On the simple side benchmarks for raw spawn performance and raw message sending are great. The "more complicated" benchmarks should be simple programs that are frequently benchmarked in other parallelism systems. Parfib, k-means, etc.

[toddaaro]

Currently we have a spawn benchmark, a parfib benchmark, and a ping-pong message passing benchmark. A start but we need expansion. Benchmarks targeted at the runtime are in the bench folder, and names start with "rt-".

[thestinger]

There are a lot of benchmarks now, and since green threads are being removed as part of #17325 I don't think there's going to be much to do that's interesting. It will just perform as well as the native building blocks provided by the platform.