This is a problem with packaging, when the binary is compiled on
a machine supporting newer SSE extensions than the box where the
package is installed. Moreover, CFLAGS decisions should be up to
maintainer or the one building the software.

This is O(N) on both insert/search, but since the ACL lists are
several items long (usually), it is not an issue.
The lists are sorted from the longest prefix to the shortest, so
the first match is guaranteed to be longest prefix match.

Fixed a bug where NSID couldn't be switched off on reload, causing
malformed packets and/or extra bytes.
Since it's protected by RCU, it needs to be created on each server
reload and the old one should be freed after the RCU is synchronized.
Also removed rest of obsolete code and unified EDNS constants.

refs #137

Problem was the function is copied from another one which
always decrements the arcount and trims the TSIG RR. But with
aggregated responses, this should be stripped not from first,
but from the packet actually containing the TSIG.

refs #122

refs #118

refs #118

Either set it to 0 or comment out the statement.

Main changes:
* 'system.hostname' config option is deprecated
* 'system.identity' is used for 'id.server' and 'hostname.bind'
* 'system.version' is used for 'version.server' and 'version.bind'
* In addition, config options (and NSID) support boolean on|off
* When on, the automatic default is filled on config load:
  * FQDN hostname (or at least hostname) for NSID and system.identity
  * Version string for system.version

The config options are still turned off by default.
refs #113

There was also a small error when a TCP master socket was removed,
the next one was skipped regardless of it's activity level.

If previous xfer socket was disconnected, the 'ret' variable wasn't cleared
when the current socket is inactive. The outcome of this is that the
innocent transfer could be discarded.

Thanks to Jonathan Hoppe from Netriplex for discovering this.

refs #111

* xfrin_add_new_node and ddns_add_new_node were duplicates
  - knot_node_set_zone() was called in the _add functions anyway
* merged checks for NSEC3-related RR to knot_rrset_is_nsec3rel()
  - the empty RR check was pointless, as even if empty, the node
    shouldn't be searched in normal tree