Fix when we switch synchronous_standby_names to '*'.

[DimCitus]

Because we assign both SECONDARY and PRIMARY in the same monitor
ProceedGroupState call, we might end up with the primary fetching its new
synchronous_standby_names from the monitor before when the secondary has
made it to its new state.

With the previous coding, the primary would then retrieve '', meaning sync
replication is disabled, when what's needed is '*'.

We add some needed test coverage from the situation and fix it in the
monitor by returning '*' as soon as the other node is assigned SECONDARY
rather than only after it has reported the state.

[jchampio]

With this PR branch, I still get intermittent failures as in #481:

======================================================================
FAIL: test_multi_standbys.test_014_update_standby_names_when_adding_and_removing_a_standby
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/usr/lib/python3/dist-packages/nose/case.py", line 197, in runTest
    self.test(*self.arg)
  File "/usr/src/pg_auto_failover/tests/test_multi_standbys.py", line 245, in test_014_update_standby_names_when_adding_and_removing_a_standby
    eq_(node1.get_synchronous_standby_names(),
AssertionError: '' != '*'

I'm not sure that the work here is related, because test_014 is missing calls to wait_until_state().

[onderkalaci]

This looks good, and I verified that the race condition is gone. (And, wow it was almost always setting the sync standbys to `` before the patch)

With this PR branch, I still get intermittent failures as in #481:

As we talked in a private chat, we realized that we might need to find the reason for #481 separately as that's multiple standbys are not going through this code-path

[JelteF]

As discussed through chat, some test ordering should still be changed. Other than that this looks good to me.