Fix ICE when ABI computation fails for unreachable/no-codegen instances

[rynewang]

Summary

Fix ICE when ABI computation fails for monomorphized instances that never reach codegen.

Problem

check_instance_abi in abi_check.rs uses delayed_bug for ABI computation failures, assuming codegen will report the actual error. However, if the instance is in unreachable code or compiled with -Zno-codegen, codegen never processes it and the error goes unreported, causing the delayed_bug to fire as an ICE.

Reproducer (#152204)

#![feature(portable_simd)]
fn main() {
    if false {
        let _ = core::simd::Simd::<u8, 256>::splat(0);
    }
}

Reproducer (#149156)

#![crate_type = "lib"]
struct Struct([u8; 0xffff_ffff_ffff_ffff]);
pub fn function(value: Struct) -> u8 { value.0[0] }
// compile with: rustc -Zno-codegen

Fix

Replace delayed_bug with emit_err in check_instance_abi to report the layout error directly rather than deferring to codegen. FnAbiError already implements Diagnostic, so the existing error infrastructure handles all variants (SizeOverflow, InvalidSimd, etc.).

After this change:

• #152204 emits: error: the SIMD type 'Simd<u8, 256>' has more elements than the limit 64
• #149156 emits: error: values of the type '[u8; 18446744073709551615]' are too big for the target architecture
Includes regression test.

Fixes #152204
Fixes #149156

[rustbot]

r? @JonathanBrouwer

rustbot has assigned @JonathanBrouwer.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: compiler
• compiler expanded to 21 candidates
• Random selection from 12 candidates

[Bryntet]

The comments in your files are overly verbose IMO

[rynewang]

Fixed the comments

[workingjubilee]

r? @workingjubilee