Biased reference counting for Rust - Thread-safe like an Arc, but as fast as a Rc for the biased thread.

This is the same reference counting technique that Swift and free-threaded Python use. I wrote this a while ago but hadn't published it until today. The performance in the single-threaded case isn't quite as good as Rc, but it's much better than Arc!

I lost my benchmark results, but you can see the improvement for yourself with cargo bench.

The code passes miri and has good test coverage, but please let me know if you see any other bugs.