"stabilize never type" T-types FCP

[lcnr]

We forgot to T-types FCP #155499. See that PR for more detail.

As discussed on zulip:

My specific recommendation here would be that once we get a complete sense of the type system hacks that we're landing to support the stabilization, that we do a T-types FCP on the PR to confirm we're good.

This PR has three changes which make it worthy of a Types FCP

---

Allow users to name ! directly. It was already possible to use ! outside of function return types by using the following pattern:

trait Func {
    type Ret;
}
impl<F: FnOnce() -> R, R> Func for F {
    type Ret = R;
}
type Never = <fn() -> ! as Func>::Ret;

However, I remember us generally discouraging users from writing such impls. I don't think there are any significant complications from users being able to implement traits for ! more widely. So this seems fine to me.

---

This PR also changes the "never type fallback" to always fall-back to ! instead of () in old editions: 6656056. See the crater results for that change in #155499 (comment). This affects any inference variable resulting from a never-to-any coercion which we apply whenever we're coercing an value expression with type !, see #155499 (comment).

---

Finally, we're setting Infallible to be a type alias for ! instead of an enum without any variants. This was done in #155924. While this change is T-libs, we did 3 Types FCPs to minimize its impact.

@WaffleLapkin mentioned 4 kinds of breakage in #155924 (comment):

1. Calling a method on a result from a coercion of never to any currently results in an error (see Inconsistent method lookup on never type #143349)
   • fixed by Fix trait method resolution on an adjusted never type #156047
2. It is possible to use hacks to avoid stability checking for the never type, thus being able to use it on stable. There are a few crates which do this and implement the same trait for both never and Infallible (or also crates which enable the nightly feature).
   • this breakage has not been fixed and there are 9 root causes and a few more dependencies which break because of that. This is fine and not a T-types issue :>
3. standard library contains a reservation impl, which forbids Tracking issue for reserved impl impl<T> From<!> for T #64715 From<!> impls. After making Infallible = !, this reservation impl can conflict with existing implementations for Infallible - This breaks 14 crates total (including reverse-dependencies of broken crates)
   • fixed by Remove From<!> for T *reservation* impl #160705
4. There is some weirdness with how coercions work, which I don't yet understand
   • fixed by consider subtyping when checking if an infer var is sized #157820

[lcnr]

All three of these changes seem acceptable to me. The remaining breakage is acceptable to me and the cost of the one temporary FCW for method calls on ! is not that high.

@rfcbot fcp merge types

[rust-rfcbot]

@lcnr has proposed to merge this. The next step is review by the rest of the tagged team members:

• @BoxyUwU
• @jackh726
• @lcnr
• @oli-obk
• @spastorino

No concerns currently listed.

Once a majority of reviewers approve (and at most 2 approvals are outstanding), this will enter its final comment period. If you spot a major issue that hasn't been raised at any point in this process, please speak up!

See this document for info about what commands tagged team members can give me.

[lcnr]

The only temporary hack is #156047 which has been FCP'd separately. The question is only "does all of this together seem acceptable"