Convert STORE_BLK into STORIND for SIMD

[EgorBo]

Closes #116258

       xor      eax, eax
       mov      dword ptr [rdi+0x08], eax
       vxorps   xmm0, xmm0, xmm0
-      vmovdqu  xmmword ptr [rdi+0x10], xmm0
-      vxorps   xmm0, xmm0, xmm0
+      vmovdqu  ymmword ptr [rdi+0x10], ymm0

STORE_IND<SIMD*> is more coalescing-friendly

Some minor diffs

[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

Adds a hardware‐intrinsics path to transform zero‐init and const‐init SIMD stores from STORE_BLOCK to STORE_IND, using a vector constant node for better coalescing.

• Introduces FEATURE_HW_INTRINSICS guard with const‐value handling
• Builds a GenTreeVecCon from a byte‐filled simd_t and replaces the original init node
• Changes the block operation to GT_STOREIND and invokes common lowering

[EgorBo]

@jakobbotsch @dotnet/jit-contrib small change

if we have GT_STORE_BLK<TYP_STRUCT> that is initialized with CNS_INT, we can replace it with GT_STORIND<SIMD>. The latter is more coalescing friendly

[EgorBo]

hm.. oops, some unexpected diffs on arm64, let me take a look

UPD: I've just disabled it on arm64, SIMD stores can't be merged on arm64 anyway + arm64 has zero-reg

[jakobbotsch]

A couple of nits, and the TODO-CQ in the function header can be removed now I think. Otherwise LGTM.