fix(mesh-phase5): telemetry hub — plain Dictionary under gate, concurrency stress test, priming/eviction notes

Claude-Session: https://claude.ai/code/session_01GASWkNEi68FSCtvr6rLoEW

src/Server/ZB.MOM.WW.OtOpcUa.Runtime/Telemetry/TelemetryLocalHub.cs

@@ -28,10 +28,18 @@ namespace ZB.MOM.WW.OtOpcUa.Runtime.Telemetry;
 public sealed class TelemetryLocalHub : ITelemetryLocalHub
 {
     private readonly object _gate = new();
-    private readonly ConcurrentDictionary<Guid, Channel<TelemetryItem>> _subscribers = new();
+
+    // Plain Dictionary, NOT ConcurrentDictionary: every access is already under _gate, and
+    // ConcurrentDictionary.Values would materialize a fresh List (and take its own internal lock) on
+    // every Emit — the hot node-actor-thread path. Under the gate a plain Dictionary is safe and
+    // allocation-free to iterate.
+    private readonly Dictionary<Guid, Channel<TelemetryItem>> _subscribers = new();
 
     // Snapshot last-value caches. Only ever mutated under _gate (kept ConcurrentDictionary so a
     // Subscribe reading them under the gate is trivially safe even against any future lock-free read).
+    // TODO(mesh-phase5+): evict snapshot cache entries on driver-instance removal (needs a
+    // driver-lifecycle hook); pre-production, accepted for now. Until then a decommissioned driver
+    // instance's last-known Health/Resilience state replays to new subscribers forever.
     private readonly ConcurrentDictionary<string, TelemetryItem.Health> _healthCache = new();
     private readonly ConcurrentDictionary<(string InstanceId, string HostName), TelemetryItem.Resilience> _resilienceCache = new();
 
@@ -91,7 +99,7 @@ public sealed class TelemetryLocalHub : ITelemetryLocalHub
     {
         lock (_gate)
         {
-            if (_subscribers.TryRemove(id, out var channel))
+            if (_subscribers.Remove(id, out var channel))
                 channel.Writer.TryComplete();
         }
     }