Fix read_cycle_counter intrinsic on arm64

Running this code with the hard-coded x9 register results in not 
returning the cycle count, but some other value.

This code (using $0 as the register), works on arm64 Linux and Darwin.

src/llvm_backend_proc.cpp

@@ -2115,7 +2115,7 @@ gb_internal lbValue lb_build_builtin_proc(lbProcedure *p, Ast *expr, TypeAndValu
 			if (build_context.metrics.arch == TargetArch_arm64) {
 				LLVMTypeRef func_type = LLVMFunctionType(LLVMInt64TypeInContext(p->module->ctx), nullptr, 0, false);
 				bool has_side_effects = false;
-				LLVMValueRef the_asm = llvm_get_inline_asm(func_type, str_lit("mrs x9, cntvct_el0"), str_lit("=r"), has_side_effects);
+				LLVMValueRef the_asm = llvm_get_inline_asm(func_type, str_lit("mrs $0, cntvct_el0"), str_lit("=r"), has_side_effects);
 				GB_ASSERT(the_asm != nullptr);
 				res.value = LLVMBuildCall2(p->builder, func_type, the_asm, nullptr, 0, "");
 			} else {