Up to now, pg_regress --config-auth had a hard-wired assumption
that the target cluster uses the default bootstrap superuser name.
pg_dump's 010_dump_connstr.pl TAP test uses non-default superuser
names, and was klugily getting around the restriction by listing
the desired superuser name as a role to "create".  This is pretty
confusing (or at least, it confused me).  Let's make it clearer by
allowing --config-auth mode to be told the bootstrap superuser name.
Repurpose the existing --user switch for that, since it has no
other function in --config-auth mode.

Per buildfarm.  I don't have an environment at hand in which I can
test this fix, but the buildfarm should soon show if it works.

Discussion: https://postgr.es/m/3142.1561840611@sss.pgh.pa.us