OK. It looks good to me. The only thing which surprised me that in test case we see unnecessary sort in remote query. However, it's explained by selected costs and STD_FUZZ_FACTOR, so that sorted path has essentially the same cost as non-sorted one according to compare_path_costs_fuzzily().
Yeah, and meanwhile the sorted path is considered to have better pathkeys than the non-sorted one. So it wins in add_path().
