Currently we have two separate structures to deal with Ethereum Node Records, a "generic" one, named NodeRecord, that allocates to keep the pairs, and a specific one with the ones we understand, named NodeRecordPairs. I think we should move to using only the latter. Our own local_node_record currently uses the former, cloning a vector for every message. We should always use NodeRecordPairs, ignoring fields we don't know, like Geth does, and rejecting peers with invalid values in the ones we do know.