docs+test(site-runtime): Lever 1 final-review polish — document that the Good reset does not release an in-flight request and that the guard is inert once interval ≥ ceiling; FlappingGood asserts the retry landed during the burst instead of timing Thread.Sleep; NativeAlarmActor (same unguarded amplifier) + cancel-race residue recorded as follow-ups

Claude-Session: https://claude.ai/code/session_019ZxXT5sxZs5XrtdwYzJDPV

tests/ZB.MOM.WW.ScadaBridge.SiteRuntime.Tests/Actors/InstanceActorTests.cs

@@ -1466,7 +1466,6 @@ public class InstanceActorTests : TestKit, IDisposable
         instance.Tell(new SubscribeTagsResponse(req.CorrelationId, req.InstanceUniqueName,
             false, "busy", DateTimeOffset.UtcNow), dclProbe.Ref);
 
-        var sw = System.Diagnostics.Stopwatch.StartNew();
         // Twelve Goods over ~1.2 s — a naive re-arm-from-now on each would push the tick to ≥1.5 s.
         for (var i = 0; i < 12; i++)
         {
@@ -1474,9 +1473,9 @@ public class InstanceActorTests : TestKit, IDisposable
             Thread.Sleep(100);
         }
 
-        // The probe already holds the retry (it landed ~300 ms after the first Good) by the time the burst ends.
-        dclProbe.ExpectMsg<SubscribeTagsRequest>(TimeSpan.FromSeconds(2));
-        Assert.True(sw.Elapsed < TimeSpan.FromMilliseconds(1400), $"retry took {sw.Elapsed} — flapping Good deferred it");
+        // The retry must have landed DURING the burst (~300 ms after the first Good), so the probe
+        // already holds it — assert that directly rather than timing the burst's Thread.Sleeps.
+        dclProbe.ExpectMsg<SubscribeTagsRequest>(TimeSpan.FromMilliseconds(50));
     }
 
     [Fact]