multi-master and zone refresh

This problem applies mainly to the 1.6.

The multi-master handling currently makes little sense. If multiple masters are defined for a zone, the server will try to fetch the zone from the first remote. If the SOA query fails, the server rotates the masters' addresses and goes to sleep. After SOA-retry period, the zone refresh is retried with the new remote address.

Therefore, if the first master is not available, the zone refresh is delayed significantly. The problem is even multiplied, if a group is used instead of a single remote. The group is just an expansion of multiple addresses. So if the group contains both IPv4 and IPv6 address of one server, and the server is temporarily down, it takes 2 * SOA-retry time just to get to the second server.

NSD tries all the servers in the list of masters, one by one. And fetches the zone from the first server available.

BIND does the same, except that it prefers the server from which the NOTIFY was received.

/reported by @anandb