perf(tunnel-node): event-driven backpressure, atomic settle, async udpgw DNS

High priority:
- #1 Replace reader_task busy-poll (50ms) with drain_notify Notify —
  reader now wakes only when bytes are consumed, cutting CPU waste
  under sustained backpressure (fast upstream / slow Apps Script).
- #4 Add buf_len / pkt_count atomics so the straggler settle loop
  reads buffer sizes lock-free (was 10k+ mutex acquisitions per batch).
- #7 Make DstAddr::to_socket_addr async, use tokio::net::lookup_host
  instead of blocking std::net::ToSocketAddrs inside async tasks.

Medium priority:
- #5 Collect stale sessions under global lock but abort them after
  releasing it — other handlers aren't blocked during task abort.
- #8 Cap udpgw accumulation buffer at 256 KiB; clear on overflow to
  prevent OOM from repeated corrupt frames.

Low priority:
- Add sid_short(&str) helper, deduplicate nginx-404 DECOY_404_BODY
  const, remove redundant drop(packets), suppress dead_code on
  test-only pub consts.

Author: brightening-eyes