Rayon is underutilizing threads

[denehoffman]

This is a problem I figured I'd run into. The overhead for creating and moving new threads is greater than the performance gain. I believe this might depend on dataset size, but I think a future version should play around with thread pool settings to optimize this. Here's the current performance vs. number of threads:

The uptick at the end is probably just because this benchmark isn't isolated, and whatever processes are currently running on my computer are interfering with the cores that laddu wants to use.