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

Refs #102

     - Inserting DNAMEs from out-of-zone RRs into tmp DNAME table messed
things up when we free'd the RRSet, it's better to check before
inserting into table (one could rollback the table, but there are some
corner cases not worth solving)

Refs #102

refs #100

Refs #102

- RFC 2136 states that the server should return the 'NOTZONE' error
  code, which it is correcly doing now, so there's no need to change it.
- Minor change only, since the actual check is somewhere before this code
  executes, but I've left the new_node functions as it was, since we
  may wish to allow some malformations later.

Refs #102

Refs #102

Refs #102

- Added an error handling function
- Changed some static functions a bit - they now return error codes
  instead of NULL

Refs #102

 - This out-of-zone check caused the whole transfer to fail. IXFR needs
   to be changed to, reverting to original master (I'd just delete the
master branch to rewind, but it is locked, but oh well)

Refs #102

- Server would refuse SOA records this way and transfers would fail.

Refs #102, fixes Coverity #1049750

Refs #102

refs #102

fixes #99

Both do not interrupt any I/O, just change thread state.
This is because it is not required in most cases unless you use
infinite timeouts in I/O ops.

refs #91

refs #91

Because it was used also for thread wakeup when it isn't needed,
the only limitation is not to use infinite poll loops.
Before it was removed, it broke perfectly good transfers with EINTR.

refs #91

Like example.com..zone

This makes partial-closed connections discovery faster than waiting
full interval for the watchdog sweep.

refs #91