tests-extra: ixfr/master_pin2 -- align the servers at the start

Notice:
Both masters send their NOTIFY's to the slave. The slave makes
refresh from the last master which it received NOTIFY from
- it's the current state of Knot DNS design. If a NOTIFY
for lower serial arrives close enough after the previous NOTIFY
for a higher serial from another master, before the slave made
refresh from the first master, the first NOTIFY "knowledge" gets
"overwritten" by the second one with lower serial, resulting
in refresh from the lagging behind master and a rare test failure.

tests-extra/tests/ixfr/master_pin2/test.py

@@ -52,11 +52,26 @@ for m in [ masterA, masterB ]:
     m.dnssec(zones).enable = True
     m.conf_zone(zones).journal_content = "all"
 
+masterA.conf_zone(zones).notify_delay = "0"  # Disable even the small random value tests use.
+masterB.conf_zone(zones).notify_delay = "0"  #
 masterB.conf_zone(zones).serial_modulo = str(-PIN)
 
 t.start()
 
+ma_serials0 = masterA.zones_wait(zones)
+mb_serials0 = masterB.zones_wait(zones)
+slave.zones_wait(zones)
+
+# Align the masters before the testing starts.
+slave.ctl("zone-freeze", wait=True)
 serials0 = slave.zones_wait(zones)
+masterA.ctl("zone-sign")
+masterB.ctl("zone-sign")
+masterA.zones_wait(zones, ma_serials0)
+masterB.zones_wait(zones, mb_serials0)
+slave.ctl("zone-thaw")
+
+serials0 = slave.zones_wait(zones, serials0)
 
 for m in [ masterA, masterB ]:
     up = m.update(zone)
@@ -79,6 +94,7 @@ masterA.ctl("zone-thaw")
 
 t.sleep(4)
 
+# May fail occasionally, for details see this commit's message.
 check_cur(masterA, masterB)
 
 RUNNING = False