JIT: Coalesce stores from struct copies

[EgorBo]

Allows existing store coalescing to merge non-GC field stores produced by struct copies.

Int128 _fld;

void Foo1(out Int128 x) => x = new Int128(2L, 1L);

void Foo2() => _fld = 123456789;

 Foo1:
- mov      qword ptr [rdx], 1
- mov      qword ptr [rdx+0x08], 2
+ vmovups  xmm0, xmmword ptr [reloc @RWD00]
+ vmovups  xmmword ptr [rdx], xmm0

 Foo2:
- mov      qword ptr [rcx+0x08], 0x75BCD15
- xor      eax, eax
- mov      qword ptr [rcx+0x10], rax
+ vmovss   xmm0, dword ptr [reloc @RWD00]
+ vmovups  xmmword ptr [rcx+0x08], xmm0

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull request overview

This PR updates JIT copy-block morphing so that when a struct copy is expanded into multiple field stores, the non-GC field STOREIND nodes are marked with GTF_IND_ALLOW_NON_ATOMIC. This enables existing lowering-time store coalescing to legally merge adjacent non-volatile stores (e.g., two 8-byte stores into one 16-byte SIMD store) without being blocked by atomicity constraints.

Changes:

• In MorphCopyBlockHelper::CopyFieldByField, mark per-field indirect stores as GTF_IND_ALLOW_NON_ATOMIC when they originate from a copyBlk and the field type is not GC (!varTypeIsGC(srcType)).
• This allows Lowering’s store coalescing (LowerStoreCoalescing / LowerCheckCoalescedStoreAtomicity) to treat these adjacent stores as eligible for wider, potentially unaligned, non-atomic combined stores.

[EgorBo]

@MihuBot -nuget