[AArch64] Implement vector float->double widening and double->float narrowing.