Making RWLock::new const fn

[elichai]

Related #66806
The only problem I saw was the fact that sys::RWLock was on the heap, but I couldn't find any reason why so I removed the Box.

If there's a reason why it was boxed that I missed please tell me.

I also didn't see any good reason to put this behind a feature, so if this will be accepted as is it will be instant stable.

[rust-highfive]

r? @KodrAus

(rust_highfive has picked a reviewer for you, use r? to override)

[Mark-Simulacrum]

r? @alexcrichton

But I believe we can't remove the box here -- there's a comment on the mutex impl that indicates that the underlying system primitives can't move after creation (https://github.com/rust-lang/rust/blob/master/src/libstd/sync/mutex.rs#L112-L116) -- I suspect the same is true of rwlock, we just aren't documenting that today (though probably should do so).

[alexcrichton]

Thanks for the PR! @Mark-Simulacrum is right though in that the system primitives are required to be boxed up here for safety, so technically exposing the underlying API as safe is actually incorrect.