Make the server handler tests more stable

The tests were randomly failing due to the number of received change
notifications in the test was sometimes not matching the expected number.
This was because sometimes the thread with the FakeClient would race
with the main test thread by starting replaying the test scenario before
the main thread has acquired the NotificationController and installed
QSignalSpy to it. In this case the QSignalSpy would have missed the
notification.

The fix is to split the FakeClient start up - first phase is for the
FakeClient to connect to the server, then the test acquires the
NotificationController and installes QSignalSpy on it and only after
that it instructs the FakeClient to start replaying the test scenario.