LLVM target triple for Graviton2 CPUs

Hmmm it should just follow the LLVM’s convention