SR tests build each Software Release from an HTTP URL, so a profile
assuming ${:_profile_base_location_} is a filesystem directory is caught
the way production catches it. A per-test-process server would be
fragile: it dies with its test, so a later test's `slapos node software
--all` re-installs an earlier, still-supplied release from a dead URL and
cascades unrelated failures. A single supervised server in the instance
stays up for the whole session instead.