Fix RwLock fallback implementation

[seritools]

An RwLock fallback implementatoin cannot be done with a mutex, as it prevents multiple read locks e.g. from the same thread to happen, or prevents co-dependent threads that all should have read locks from progressing together. In fact, even the simple example in the Rust std docs for RwLock will panic immediately.

Split off from #32 (comment)

Interesting links:

• https://eli.thegreenplace.net/2019/implementing-reader-writer-locks/
• https://old.reddit.com/r/cpp/comments/1b55686/maybe_possible_bug_in_stdshared_mutex_on_windows/kt7ovpx/

[seritools]

This one might be very easy to vendor in: https://github.com/cmnord/pflock nvm, does not suspend threads when waiting

[seritools]

https://justine.lol/mutex/

[seritools]

Newer Rust versions have their own SRWLOCK-like implementation that doesn't rely on OS primitives beyond a thread parker implementation. We can bring back the old generic thread parker implementation (as already done in the 1.76 rust9x version) and enable a proper RwLock impl that way 🚀

[seritools]

Fixed on the 1.84 branch (in development)