feat: per-listener latency tracking

Each listener now tracks chain setup latency independently via a
dict[str, LatencyTracker] on Metrics. The global aggregate stays for
summary output. /status embeds per-listener latency on each listener
entry; /metrics includes a listener_latency map keyed by host:port.

tests/test_metrics.py

@@ -136,6 +136,23 @@ class TestMetrics:
         assert "p50=" not in s
         assert "p95=" not in s
 
+    def test_listener_latency(self):
+        m = Metrics()
+        m.get_listener_latency("0.0.0.0:1080").record(0.5)
+        m.get_listener_latency("0.0.0.0:1080").record(0.6)
+        m.get_listener_latency("0.0.0.0:1081").record(0.1)
+        d = m.to_dict()
+        assert "listener_latency" in d
+        assert "0.0.0.0:1080" in d["listener_latency"]
+        assert "0.0.0.0:1081" in d["listener_latency"]
+        assert d["listener_latency"]["0.0.0.0:1080"]["count"] == 2
+        assert d["listener_latency"]["0.0.0.0:1081"]["count"] == 1
+
+    def test_listener_latency_empty(self):
+        m = Metrics()
+        d = m.to_dict()
+        assert d["listener_latency"] == {}
+
 
 # -- _human_bytes ------------------------------------------------------------