On rotational devices (HDD) processing direntries is much faster when
doing so sorted by i-node for the dir-entries. This is an old
optimization (perhaps mu <= 1.6 or so?) that was implemented yet after
indexing changed, likely because my systems use SDDs instead!
But, let's restore that optimization; the sorting is fast enough that we
don't care for SDDs; on HDD it should be quite a bit faster.