Optimize NDR uop_CALL_INSTRUCTION_FUNC by loading the fetchdat in one uOP instead of two

Another 3-5% NDR improvement noticed on the WOLF3D MAPEDIT idle loop as a result.

src/codegen_new/codegen_backend_arm64_uops.c

@@ -218,6 +218,7 @@ codegen_CALL_FUNC_RESULT(codeblock_t *block, uop_t *uop)
 static int
 codegen_CALL_INSTRUCTION_FUNC(codeblock_t *block, uop_t *uop)
 {
+    host_arm64_mov_imm(block, REG_ARG0, uop->imm_data);
     host_arm64_call(block, uop->p);
     host_arm64_CBNZ(block, REG_X0, (uintptr_t) codegen_exit_rout);