Race condition in MultiThreadedExecutor with intra_process communication resulting in dequeuing ring buffer before enqueuing

[brawner]

There is a race condition in the MultiThreadedExecutor that happens during the test_multithreahed.multi_consumer_intra_process test in test_rclcpp. The intra_process_buffer's ring_buffer_implementation has dequeue() called without any data. It may be that some process is calling dequeue() right after another has checked has_data().

This bug results in a flakey test with test_rclcpp, but it's fairly reliable to reproduce with sufficient repeat-until-fail iterations.

https://ci.ros2.org/view/nightly/job/nightly_linux_repeated/1780/testReport/(root)/projectroot/gtest_multithreaded__rmw_fastrtps_cpp/
https://ci.ros2.org/job/ci_windows/9412/testReport/