It appears dcae5fac forgot to add it to
pg_isolation_regress_installcheck, while it was added to
pg_regress_installcheck.  It seems to so far have escaped notice,
because buildfarm animals requiring it, didn't actually use
pg_isolation_regress_installcheck anywhere - that changed with
60f826c5, triggering failures on narwhal and frogmouth.

I've decided to not, for now at least, backpatch this, because the
relevant invocations look quite different in the back branches.  Seems
quite possible that we'll want to backport 60f826c5 as a whole if
it proves stable.

Discussion: https://postgr.es/m/20170315174003.3dyl4teashdwgblh@alap3.anarazel.de