Feature: Bridge coverage boost for uncovered lines in bridge.py
  As a developer
  I want thorough tests for remaining uncovered lines in bridge.py
  So that code coverage reaches its maximum

  # Targets uncovered lines:
  #   Line 64:  early return from cleanup_tasks_async when no tasks exist
  #   Lines 79-83: pending tasks that do not complete within timeout
  #   Lines 89-93: late tasks added to _active_tasks during cleanup await
  #   Line 149: MESSAGE_ROUTER node with "rules" in create_graph_from_config
  #   Lines 282-290: execute_node inner coroutine in _create_node_operator

  Scenario: cleanup_tasks_async returns immediately when no tasks are active
    Given a bridge with an empty active tasks set
    When I call cleanup_tasks_async on the idle bridge
    Then the async cleanup should complete without error
    And the bridge active task set should remain empty

  Scenario: cleanup_tasks_async logs pending tasks that exceed timeout
    Given a bridge with a task that ignores cancellation
    When I call cleanup_tasks_async with a very short timeout
    Then pending tasks should be re-cancelled after the timeout
    And the bridge active task set should be cleared after timeout cleanup

  Scenario: cleanup_tasks_async cancels late tasks added during await
    Given a bridge with a task whose done-callback spawns a late task
    When I call cleanup_tasks_async allowing callback to fire
    Then the late task should be cancelled during cleanup
    And the bridge active task set should be cleared after late-task cleanup

  Scenario: create_graph_from_config sets rules metadata for MESSAGE_ROUTER nodes
    Given a bridge ready to create graphs from config
    When I create a graph with a MESSAGE_ROUTER node containing rules
    Then the node config metadata should contain the routing rules

  Scenario: Node operator executes the inner coroutine and updates state
    Given a bridge with a graph containing a mock executable node
    When I pipe a message through the node operator and await the result
    Then the node execute method should have been called
    And the state manager should have been updated with the node result
    And the returned message metadata should contain the node and graph names