By splitting the `FnSig` within `TyKind::FnPtr` into `FnSigTys` and
`FnHeader`, which can be packed more efficiently. This reduces the size
of the hot `TyKind` type from 32 bytes to 24 bytes on 64-bit platforms.
This reduces peak memory usage by a few percent on some benchmarks. It
also reduces cache misses and page faults similarly, though this doesn't
translate to clear cycles or wall-time improvements on CI.
elaborate unknowable goals
A reimplemented version of #124532 affecting only the new solver. Always trying to prove super traits ends up causing a fatal overflow error in diesel, so we cannot land this in the old solver.
The following test currently does not pass coherence:
```rust
trait Super {}
trait Sub<T>: Super {}
trait Overlap<T> {}
impl<T, U: Sub<T>> Overlap<T> for U {}
impl<T> Overlap<T> for () {}
fn main() {}
```
We check whether `(): Sub<?t>` holds. This stalls with ambiguity as downstream crates may add an impl for `(): Sub<Local>`. However, its super trait bound `(): Super` cannot be implemented downstream, so this one is known not to hold.
By trying to prove that all the super bounds of a trait before adding a coherence unknowable candidate, this compiles. This is necessary to prevent breakage from enabling `-Znext-solver=coherence` (#121848), see tests/ui/coherence/super-traits/super-trait-knowable-2.rs for more details. The idea is that while there may be an impl of the trait itself we don't know about, if we're able to prove that a super trait is definitely not implemented, then that impl would also never apply/not be well-formed.
This approach is different from #124532 as it allows tests/ui/coherence/super-traits/super-trait-knowable-3.rs to compile. The approach in #124532 only elaborating the root obligations while this approach tries it for all unknowable trait goals.
r? `@compiler-errors`
