pkt: switched to allocd rr/rrinfo fields to reduce memory footprint

this comes with some extra CPU cost, however:
- parsed packets are prealloc'd based on header information, so no
  reallocs
- constructed packets are alloc'd by steps
- packet construction / copy is  cheaper now
- doesn't matter when the mm backend is mempool anyway

the rationale for this change is the use in resolver and in smaller
installations