Implement a fairer locking strategy for a new mutex type

[notgull]

This PR implements a locking strategy for the SpinMutex type that uses eventual fairness, inspired by the one used in the async-lock crate. If a lock operation spins for a long period of time, it becomes a "starving" operation. When the mutex is eventually unlocked, starving locks are given precedence over non-starving locks. This is similar to less strict version of TicketMutex, in that it reduces contention in the worst cases.

The only real downside of this strategy is that is uses an AtomicUsize instead of an AtomicBool to keep track of the number of starving operations in addition to the lock status.

[zesterer]

Thanks for implementing this! I wonder whether this should be a separate implementation with the top-level Mutex type alias pointing to it?

[notgull]

Alright, I merged it out into its own file. Do you want me to add a feature to make the main Mutex type use it as well? I'd be fine without.

[zesterer]

Thanks, this is great. I'm hoping I'll get time to review this this weekend coming. Regarding feature flags, I might do some rearranging of things after this PR has been merged, so don't worry about that.

[notgull]

Thank you!

I might do some rearranging of things after this PR has been merged

I'd like to contribute to this crate more, so let me know if you want help with this "rearranging".

[notgull]

@zesterer Have you had a chance to review this PR yet?

[zesterer]

After spending some time reading through the implementation, I'm fairly confident in the code. Just a few more minor questions (I hope you understand: this is quite a large change, if not in terms of lines of code, in terms of 'number of users across the ecosystem').

[notgull]

(You probably want to squash this, by the way.)

[zesterer]

Thanks!