We don't want this code running except on Red Hat systems, where we can
maintain it in a downstream patch if need be.

The heuristic has been wrong before (e.g. as in #57) and is not useful
on upstream systems where 1000 is the presumed minimum uid for human
users.