More efficient monitoring of fibers on virtual threads

[armanbilge]

Spinning out of #3929 (comment).

We fixed the memory leak, but monitoring fibers on virtual threads still has a single point of contention.

[djspiewak]

In general this is going to be a difficult space for us. There are probably a ton of places where we assume the absolute number of threads is small while the number of fibers is large. I agree we should try to fix this issue, but we're probably going to be playing whack-a-mole a bit.

Overall, we should probably consider where along the spectrum from "ignoring virtual threads" to "we have a dedicated runtime and fiber implementation when we detect virtual thread support" we want to fall.