Fix race condition in drain_streams_concurrently ensure block

The ensure block was checking client_disconnected before waiting for
writing_task to complete. This caused a race where the flag might still
be false when checked, even though writing_task would set it to true
moments later.

Fixed by reordering the ensure block:
1. Close queue (unblocks writing_task if waiting on dequeue)
2. Wait for writing_task (ensures flag is set if disconnect occurred)
3. Check flag and stop barrier if needed

This ensures producers are properly stopped when clients disconnect,
preventing wasted CPU cycles.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <[email protected]>

Author: ihabadham

react_on_rails_pro/lib/react_on_rails_pro/concerns/stream.rb

@@ -93,12 +93,16 @@ def drain_streams_concurrently(parent_task)
         raise e
       end
     ensure
-      # If client disconnected, stop all producer tasks to avoid wasted work
-      @async_barrier.stop if client_disconnected
-
-      # Close the queue to signal end of streaming
+      # Close the queue first to unblock writing_task (it may be waiting on dequeue)
       @main_output_queue.close
+
+      # Wait for writing_task to ensure client_disconnected flag is set
+      # before we check it (fixes race condition where ensure runs before
+      # writing_task's rescue block sets the flag)
       writing_task.wait
+
+      # If client disconnected, stop all producer tasks to avoid wasted work
+      @async_barrier.stop if client_disconnected
     end
 
     def log_client_disconnect(context, exception)