rust_port_detach is very inefficient

[brson]

This function does a busy wait until there are no outstanding references to the port. I believe this is what's causing #1922 - under valgrind it serializes threads so we end up waiting a very long time.

[brson]

I'm not sure how big a problem this is. It's obviously a big problem for valgrind, just based on the way valgrind works - callgrind shows it as taking up 50% of the running time on some benchmarks. In real multithreaded scenarios it is probably not nearly as bad. Still, it should be fixed just to reduce the cycle time on the linux bots.

[brson]

Detaching now puts tasks to sleep as necessary instead of busy waiting.