fix(cortex-engine): handle SIGPIPE by detecting closed event channel

Fixes bounty issue #1525

When cortex run is piped to a command that closes the pipe (e.g., head -n 1),
the session event channel could return a Closed error. Previously this was ignored,
causing the session to keep running. Now we detect TrySendError::Closed and properly
shut down the session.

Author: factorydroid

cortex-engine/src/session.rs

@@ -1374,8 +1374,21 @@ impl Session {
             let _ = recorder.record_event(&event);
         }
 
-        // Non-blocking send - never wait
-        let _ = self.event_tx.try_send(event);
+        // Non-blocking send
+        // If the channel is closed (receiver dropped), we should stop the session
+        match self.event_tx.try_send(event) {
+            Ok(_) => {}
+            Err(async_channel::TrySendError::Full(_)) => {
+                // Channel full - dropping event (shouldn't happen with unbounded)
+                tracing::warn!("Event channel full, dropping event");
+            }
+            Err(async_channel::TrySendError::Closed(_)) => {
+                // Channel closed - receiver is gone (e.g. pipe broken)
+                tracing::info!("Event channel closed, stopping session");
+                self.running = false;
+                self.cancelled.store(true, Ordering::SeqCst);
+            }
+        }
     }
 
     /// Resume a session from a rollout file.