fix(18): stop ends everything — pills clear, pump/prep failures roll the session back

Creator ruling 2026-09-01 (stuck REC pill after the failed first recording):
- StopStream clears RecordPillOn + OnAirPillOn — intent resets with reality.
- OnFramePumpFailed: dispatcher-marshalled FULL rollback via StopStream (was:
  toast + Error-status only when live — record-only sessions zombied with
  IsRecording=true over a dead encoder; the 20:34 attempt proved it). Toast copy
  says 'Recording stopped' vs 'stream pipeline stopped'.
- Same zombie class in the three go-live prep failure branches: StreamStatus.Error
  limbo replaced by StopStream() (audio loop + recording + pills unwind; a created
  broadcast still gets the close-out).
- AudioMixer.StopLive made explicitly idempotent (Stop() twice, rollback paths that
  never reached StartLive).
- Seams: OnFramePumpFailed + IsRecording setter internal (InternalsVisibleTo; test
  pattern mirrors LayoutPathOverride).
ONE integration test: SessionTeardownTests (real window + temp DB: pump death with
lit pill -> no zombie, no End button, Offline). AudioPipelineTests confirmed to hang
STANDALONE (pre-existing, the declared-known audio class) — ai.md test-count
paragraph corrected to stop claiming a suite total that cannot currently be measured.

ViewModels/MainViewModel.Streaming.Operations.cs

@@ -112,7 +112,7 @@ public partial class MainViewModel : ViewModelBase
                 AppLog.Write("Reusable stream unavailable; check the OAuth session");
                 _notifications.Error("Couldn't go live",
                     "The reusable stream is unavailable — check the YouTube sign-in and try again.");
-                StreamStatus = StreamStatus.Error;
+                StopStream(); // full rollback — audio loop + recording + pills, no Error-status zombie
                 return;
             }
             _reusableStream = stream;
@@ -125,7 +125,7 @@ public partial class MainViewModel : ViewModelBase
             {
                 AppLog.Write("Broadcast creation failed; check the OAuth session");
                 _notifications.Error("Couldn't go live", "YouTube rejected the broadcast — check the sign-in and try again.");
-                StreamStatus = StreamStatus.Error;
+                StopStream(); // full rollback
                 return;
             }
             AppLog.Write($"Broadcast created: {_currentBroadcastId}");
@@ -163,7 +163,7 @@ public partial class MainViewModel : ViewModelBase
         {
             AppLog.Write($"Go-live preparation failed: {ex.Message}");
             _notifications.Error("Couldn't go live", ex.Message);
-            StreamStatus = StreamStatus.Error;
+            StopStream(); // full rollback — a created broadcast (if any) gets the close-out
             return;
         }
 
@@ -227,6 +227,11 @@ public partial class MainViewModel : ViewModelBase
 
         StreamStatus = StreamStatus.Offline;
         IsRecording = false;
+        // Stop ends EVERYTHING (creator ruling 2026-09-01): a lit pill with no session
+        // behind it is a lie — the stuck REC pill after the failed 2026-09-01 recording
+        // attempt was the report. Intent resets with reality; re-arming is one click.
+        RecordPillOn = false;
+        OnAirPillOn = false;
         WindowTitle = "LlamaCasty";
         ResetHealth(StreamStatus.Offline);
         // Stop chat polling first (TASK 9.4) — no more messages arriving while
@@ -290,11 +295,27 @@ public partial class MainViewModel : ViewModelBase
         };
     }
 
-    private void OnFramePumpFailed(object? sender, string message)
+    // internal: test seam (mirrors LayoutPathOverride) — the rollback path is
+    // unobservable without simulating a pump death; ytLive.Tests has InternalsVisibleTo.
+    internal void OnFramePumpFailed(object? sender, string message)
     {
+        // No zombies (2026-09-01): a dead pump means NO session — the old handler
+        // toasted and (live-only) flipped StreamStatus, leaving a record-only session
+        // with IsRecording=true and no encoder (first real launch proved it). The
+        // failure now runs the NORMAL stop path on the UI thread (pills off, audio
+        // loop closed, recording finalized, close-out if live), then explains itself.
         AppLog.Write($"Frame pump failed: {message}");
-        _notifications.Error("The stream pipeline stopped", message);
-        if (IsLive) StreamStatus = StreamStatus.Error;
+        void Handle()
+        {
+            var live = IsLive;
+            _notifications.Error(live ? "The stream pipeline stopped" : "Recording stopped", message);
+            StopStream();
+        }
+        var dispatcher = System.Windows.Application.Current?.Dispatcher;
+        if (dispatcher != null && !dispatcher.CheckAccess())
+            dispatcher.BeginInvoke(Handle);
+        else
+            Handle();
     }
 
     // TASK 4 ship step 6: the encoder's parsed health (bitrate/FPS/dropped/