nx_simd_x86_test.nx

nx_simd_x86_test.nx -- AVX2-subset SIMD smoke (excludes VMUL which requires AVX-512 VPMULLQ; VMUL with --target x86_64 errors LOUD). Same lanewise correctness vs scalar as nx_simd_test.nx but limited to ops AVX2 actually supports. Run on real x86 hardware (or qemu-x86_64 with AVX2 cpu) -- our setup compiles to native x86 + runs directly via wsl bash since we're on x86 Linux/Windows. Note on runtime: nx_simd_test imports nx_kernel_v2 + nx_log which use RISC-V syscall numbers. This bench skips those and uses the x86_64 syscall wrappers directly to keep dependencies minimal.