Cleanup of GDBusConnection (ie: dispose/finalize) is inherently racy, and
the worker threads often keep connections reffed until idles complete and so
on.  Normally this is not a problem, but while testing these races become apparent

The gdbus-connection test was failing once in five times for me with the
following message:

[stef@stef tests]$ ./gdbus-connection
/gdbus/connection/basic: OK
/gdbus/connection/life-cycle: **
GLib-GIO:ERROR:gdbus-connection.c:144:test_connection_life_cycle: assertion failed: (error != NULL)
Aborted (core dumped)

This is because the /basic test was initializing a session bus singelton,
and then the first check in /life-cycle was expecting a failure to instantiate
such a singleton. Normally /basic would unref the connection, but very often
a worker thread would have the connection reffed, and the /life-cycle
test would fail due to the non-NULL singleton, and subsequent non-failure
of g_bus_get_sync().