Add Align type with const generic controlled alignment

[beepster4096]

Added:

// core::mem
#[lang = "align_type"]
pub struct Align<const N: usize>;

Unresolved problems/questions:

• The errors for invalid alignments are terrible
• Align<N> with invalid N will compile as long as its layout is never needed (see tests\ui\layout\align-type\invalid-unused.rs) Should this happen?
• Align<0> is valid and has an alignment of 1. Is this okay?

r? scottmcm

[scottmcm]

My instinct is that it definitely shouldn't pass, but maybe it doesn't fail elegantly -- perhaps it fails in the same way that [u32; usize::MAX / 4] fails, for example.

[beepster4096]

Hm, types that are too large don't fail in these positions either

[scottmcm]

Oh, because it's just a check build or something?

If it's consistent with malformed arrays that's probably fine -- for now at least.

[scottmcm]

My instinct is no, because everything other than layout should just treat it as a normal type with a generic-dependent layout, never switch on some type flag to detect it.

[scottmcm]

Yes, same as #[repr(align(0))] and mem::Alignment::new(0).unwrap().

(And how https://doc.rust-lang.org/std/primitive.usize.html#method.is_power_of_two does not treat zero as a power of two.)

[rust-bors]

☔ The latest upstream changes (presumably #139087) made this pull request unmergeable. Please resolve the merge conflicts.