Latest upstream kernel simply ignores unexpected data without
disconnecting so trying to wait for a HCI_Disconnect will only lead to a
test timeout.

If tester_setup_failed() was called, execute post_teardown() to clean
data created in pre_setup phase. In some cases android-tester
couldn't start daemon, and setup failed, but hciemu created in
pre_setup was not removed.

In case multiple authorizations happen it can trigger a second call to
agent_authorize_service while the first is still ongoing causing all
authorization to fail.

It seems some in some version of readline rl_redisplay does not update
when rl_printf is called, so this uses rl_forced_update_display which
should force the new lines to be displayed.

The idle watch used to dispatch authorization should be removed when
waiting services to be resolved.