Reason is - we rarely get more than a few different RR
types in a node (usually <4). So it makes not much sense
to build an expensive AVL tree and allocate other stuff.
Worst case is O(n) (but probably less) which is OK for this
scenario. If we bothered to order the array on insertion,
we could get O(logN) with binary search then - no extra memory
costs.

- No cycle
- Added some rrset helper functions
- calls of function changed
- TODO: knot_rrset_rdata_get_next_dname_pointer is now obsolete, since it does the same thing

- Changed pointer comparison to semi-full dname comparison.

- function rrset_get_next_dname needed a fix.

Conflicts:
	Knot.files
	src/Makefile.am
	src/libknot/nameserver/name-server.c
	src/libknot/rdata.c
	src/libknot/updates/xfr-in.c
	src/libknot/util/descriptor.h
	src/libknot/zone/zone-diff.c
	src/zcompile/parser-util.c
	src/zcompile/zparser.y

This is because BADTIME affects wire size and we
need to know it in advance.

Needs cleanup, just to make things work for a while.

This allows trie to modify duplicated nodes.

- Needed when RRs were destroyed or copied - DNAMEs insided have to be released/retained.

Limitations:
* no dname conversion yet (wrong ordering)
* some traversals won't be implemented (probably won't be needed)
* no index building yet (slow ordered search)

The reason is we break the aliasing rules a lot,
this is not neccessarily a bad thing but we need to either
choose to adhere to the rules and add code complexity or
disable it (only gcc attempts to do it by default).