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.c

@@ -746,8 +746,7 @@ codegen_skip:
         uop_MOV_PTR(ir, IREG_ea_seg, (void *) op_ea_seg);
     if (op_ssegs != last_op_ssegs)
         uop_MOV_IMM(ir, IREG_ssegs, op_ssegs);
-    uop_LOAD_FUNC_ARG_IMM(ir, 0, fetchdat);
-    uop_CALL_INSTRUCTION_FUNC(ir, op);
+    uop_CALL_INSTRUCTION_FUNC(ir, op, fetchdat);
     codegen_flags_changed = 0;
     codegen_mark_code_present(block, cs + cpu_state.pc, 8);