When concatenating FQDNs, the first name \0 label must be cut.

All the functions using them we're remade using the new wire based
API. Not safe for non-linear names though.

refs #19

fixes #23

refs #21

refs #23

More cleanup is required though, there are a lot of duplicate functions.
Ultimately, most of the packet functions should be gone in favour of the
knot_wire_* ones.

refs #22

This makes most of the use of the MM context apart from
several places, which are to be refactored:
- constructed RRs
- QNAME

refs #22

- Per user request

cov: 1038957, 1038955

cov:1038956

Conflicts:
	doc/running.texi

Refs #37

- Compilation warning fixed
- Constant with comment added
- Signed count removed
- Added missing header guards
- Changed output of estimation

Refs #37

Caused painfully slow AXFR when RRSIGs came before the covered RRs.

Refs #37

- Changed magic constant to 1.2
- Info about total server memory consumption

Refs #37

- Added malloc overhead (sizeof(size_t) for each allocation, according to libc's malloc.c source file.
- Added minimal malloc chunk size (sizeof(size_t) * 3)
- With magic constant of 1.15, results are OK, but needs more tweaking, only tested with one set of RRs and only on a 64bit system.

Refs #37

 - Used for ahtable size estimation
 - Estimate still not precise, but getting close
 - Added malloc wrapper for trie, but allocated sizes are negligible

  Refs #37

It would otherwise clash with TSIG errors, which are required to
be inverted TSIG RCODEs.

You can disable it by specifying an empty control {} section.