Perf: investigate manual inlining in layout.rs

[CAD97]

Due to how many different times the layout code is instantiated, seemingly inconsequential changes can have noticeable impact on compile perf.

It may be beneficial to investigate some manual inlining (open coding) of code in the layout code (e.g. replacing ? and other option/result combinators with match/return) to reduce the amount of IR LLVM has to deal with.

[nnethercote]

Another idea: Layout::from_size_align does two checks. One for alignment, one for size. I think the places where it was used in #95295 only need the size check. So an alternative unsafe version of Layout::from_size_align that does just the size check might help.

[scottmcm]

mem::ValidAlign represents the alignment check in the type system, so I think it could even be safe but still only have the one check.

[CAD97]

If only tidy hadn't slowed me down I could've posted the PR before you said that 😛 (GitHub says the difference was 24 seconds)