[native_cpu] Basic/built-ins/vec_math.cpp & const_vec_math.cpp failed

[fzou1]

Describe the bug

This is for LLVM SPIRV pulldown (WW36): #19997

Test run: https://github.com/intel/llvm/actions/runs/18011712872/job/51249845552?pr=19997#logs

The reduced test is passed after the intrinsic is changed manually from

 %1 = tail call noundef <2 x float> @llvm.maximumnum.v2f32(<2 x float> %agg.tmp.sroa.0.4.vec.insert.i, <2 x float> splat (float 5.000000e+00))

to:

 %1 = tail call noundef <2 x float> @llvm.maxnum.v2f32(<2 x float> %agg.tmp.sroa.0.4.vec.insert.i, <2 x float> splat (float 5.000000e+00))

It's probably related to PR: llvm/llvm-project#128506

To reproduce

Environment

No response

Additional context

No response