feat: live mesh peer gossip

When a peer's hello is verified, send a gossip burst of currently
connected peers so the new peer can discover and connect to nodes it
hasn't met yet — without relying on the anchor for a full peer-join
notification for each one.

- mesh: add PendingConnect chan for gossip-discovered peer IDs
- peer: send gossip burst after hello; push unknown peers from incoming
  gossip onto PendingConnect (skip self + already-connected)
- anchor: drain PendingConnect per runOnce, initiate startOffer for
  unknown peers using the same lexicographic tiebreak as anchor join;
  drain goroutine is scoped to the runOnce lifetime to avoid using stale
  sessions after a reconnect

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

internal/anchor/client.go

@@ -100,6 +100,46 @@ func runOnce(ctx context.Context, anchorURL, netHash string, id *crypto.Identity
 
 	s := &sender{id: id, sendCh: sendCh}
 
+	// Drain gossip-discovered peers and initiate offers. Stopped when runOnce
+	// returns (via drainDone) so stale sessions from a dead connection are never reused.
+	drainDone := make(chan struct{})
+	defer close(drainDone)
+	go func() {
+		for {
+			select {
+			case pid, ok := <-m.PendingConnect:
+				if !ok {
+					return
+				}
+				mu.RLock()
+				_, already := sessions[pid]
+				mu.RUnlock()
+				if already {
+					continue
+				}
+				// Use the same lexicographic tiebreak as anchor join to avoid
+				// both sides trying to offer simultaneously.
+				if strings.Compare(string(id.PeerID()), string(pid)) <= 0 {
+					continue
+				}
+				go func(pid proto.PeerID) {
+					sess, err := startOffer(ctx, pid, id, m, s)
+					if err != nil {
+						log.Printf("anchor: gossip offer to %s: %v", pid.Short(), err)
+						return
+					}
+					mu.Lock()
+					sessions[pid] = sess
+					mu.Unlock()
+				}(pid)
+			case <-drainDone:
+				return
+			case <-ctx.Done():
+				return
+			}
+		}
+	}()
+
 	for {
 		var msg proto.AnchorMessage
 		if err := wsjson.Read(ctx, conn, &msg); err != nil {