- wrong if, no copy of NSEC3 nodes was made -> leaks.

 - node->new_node is thus obosolete.
 - since hattrie_dup made almost a full copy anyway, it's feasible to
   let the zone contents recreate the zone anew, with proper pointers.
(new node was used to set only parent pointer, others were reset anyway
in adjusting, all this will die in the new zone API, but at least like
this it's easier to maintain.)

- mm_realloc is now static, since it was only used in rr.c anyway.

- NSs in apex - some possibly problematic scenarios.
- CNAMEs - dtto

 - todo: Find file with __i386__ define or something similar
 - removed contst getter for rr rdata.
 - reused knot_rr_array_size function for size + META_SIZE like formulas

- knot_rrset_copy now returns pointer

Conflicts:
	src/knot/zone/zone-contents.c
	src/libknot/packet/pkt.c

No Default Nsec3 Tree

Two parts:

* Growable hash node size in trie
* NSEC3 tree created only if needed (NULL is treated as an empty tree)
  * I'm not sure if I did the zone-diff stuff right? :suspect:

- knot_rrset_create_rr removed, only way to insert RR into rrset now is
  when you already have RDATA, so that we never create RRSet that is not
sorted.
- rr.h doxygenated, functions moved around a bit.

 - merging can now be done directly using knot_rrs_t

http://www.ucw.cz/gitweb/?p=libucw.git;a=blob;f=ucw/bigalloc.c;hb=HEAD

Leaf hhash node size is growable (1-4 pages).
The rule is simple, when a hashtable is split the left and right
children size is clamped left and right weight to the nearest
hhash size. The sizes are chosen arbitrarily so the table size
is always a multiple of page size.