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_x86-64_uops.c

@@ -219,6 +219,11 @@ codegen_CALL_FUNC_RESULT(codeblock_t *block, uop_t *uop)
 static int
 codegen_CALL_INSTRUCTION_FUNC(codeblock_t *block, uop_t *uop)
 {
+#    if _WIN64
+    host_x86_MOV32_REG_IMM(block, REG_ECX, uop->imm_data);
+#    else
+    host_x86_MOV32_REG_IMM(block, REG_EDI, uop->imm_data);
+#    endif
     host_x86_CALL(block, uop->p);
     host_x86_TEST32_REG(block, REG_EAX, REG_EAX);
     host_x86_JNZ(block, codegen_exit_rout);