A partitioned table will always go into the hasindex = false code path.
I'm kind of thinking this patch should change that, even if the patch is not making use of the indexes, you could argue that something using set_rel_pathlist_hook might want to do something there, although, there's likely a bunch of counter arguments too.
