Without this change, "source/a/b" in non-strict mode indexes:
  (document.uid, source.uid, b.uid, 1)
  (document.uid, source.uid, a.uid, 0)
  (document.uid, source.uid, source.uid, 0)
This last line does not contain anything which cannot be found by looking
at the base category uid column, so it is wasting disk (and index) space,
costing performance.
But keep indexing a Base Category document if is it not the base category
for considered relation. It is not clear whether stopping indexation at the
first encountered Base Category document is intentional, or if recursion
should only stop when reaching the base category of considered relation.
With this change, "source/a/b" in non-strict mode indexes:
  (document.uid, source.uid, b.uid, 1)
  (document.uid, source.uid, a.uid, 0)
removing the redundancy.