neon: test for MMX/SSE instead of x86 when choosing implementation

On Elbrus we want to use MMX/SSE, so the old code was causing problems
on Elbrus since the code assumed that if MMX/SSE was enabled the Intel
types would be used.

arm/neon/types.h

@@ -174,7 +174,7 @@ SIMDE_ARM_NEON_TYPE_FLOAT_DEFINE_(64, 2, SIMDE_ALIGN_16_)
     #define SIMDE_ARM_NEON_NEED_PORTABLE_F64X1XN
     #define SIMDE_ARM_NEON_NEED_PORTABLE_F64X2XN
   #endif
-#elif defined(SIMDE_ARCH_X86) || defined(SIMDE_ARCH_AMD64)
+#elif defined(SIMDE_X86_MMX_NATIVE) || defined(SIMDE_X86_SSE_NATIVE)
   #define SIMDE_ARM_NEON_NEED_PORTABLE_F32
   #define SIMDE_ARM_NEON_NEED_PORTABLE_F64