tech-debt: Add regression tests for PRs #474-476, #480-482, #496-497, #501-502

[SebTardif]

Summary

Six PRs from Rounds 63-64 fix real bugs but have no regression tests. Each fix is a one-line match arm addition, and the bugs are in compiler internals that require specific type-system states to trigger. Tests should verify the fix prevents regression.

PRs needing regression tests

PR	Issue	Bug	Test strategy
#474	#471	ARM VFP uses triple suffix, misses Apple hard-float	tests/assembly/ or tests/codegen/ with ARM target
#475	#472	struct_lockstep_tails_raw missing ty::Pat	tests/ui/ with pattern types feature
#476	#469	ReverseMapper missing CoroutineClosure + fold_const no recurse	tests/ui/ with async closures + opaque types
#480	#479	maybe_drop_guard missing CoroutineClosure	tests/ui/ with async closure drop guards
#481	#477	simplify_nounwind_call missing CoroutineClosure + Error	tests/mir-opt/ with async closure MIR
#482	#478	describe_field_from_ty missing CoroutineClosure	tests/ui/ with async closure borrow error diagnostic

Context

These are all CoroutineClosure TyKind variant gaps (except #474 which is ARM ABI, and #475 which is ty::Pat). Async closures are stable since Rust 1.85, so test cases should compile on stable.

Acceptance criteria

• Each PR has at least one regression test that fails without the fix and passes with it
• Tests follow the patterns in rust-contrib SKILL.md "Test Writing Patterns by Area"

[SebTardif]

Update: Round 66 PRs also need regression tests

The following Round 66 PRs have source-only changes with no test files:

PR	Bug	Test difficulty
#496	EC.clear() DWO error + PassManager leak (C++ LLVM wrapper)	Hard: requires DWO file error injection; may need LLVM lit test
#497	simd_shuffle error msg uses dest_len not left_len	Medium: could add a compile-fail test with mismatched shuffle lengths
#501	s390x f128 ABI (FPR pairs)	Hard: requires s390x cross-compile or assembly output test
#502	PPC32 small struct ABI (GPR coercion)	Hard: requires PPC32 cross-compile or assembly output test

The ABI fixes (#501, #502) could potentially be tested with --emit=llvm-ir or --emit=asm tests filtered by target triple, similar to existing tests in tests/assembly/.

Expanding this issue to track tests for #496-497, #501-502 in addition to the original PRs.