Minor arm64 improvements

[kd-11]

• Do not add x64 switches to arm64 output. This happens on clang which exports both compilerx86 and compilerarm somehow.
• Do not use x86-specific intrinsics without preprocessor guards. This fixes SPU module compilation allowing SPU llvm to be used in some games.

Most games still hang after loading modules when using llvm. I haven't really checked why and tbh it's not a priority right now.

[kd-11]

You can now play arkedo 01 with SPU llvm. Yay.

[kd-11]

I should clarify, this is testing "true" arm64 (i.e linux arm64) not apple's weird arm64 impl. I am running on apple silicon here but basically side-stepping apple's nonsense and just focusing on the hardware.