My primary question is how can we test to verify all the leaks are gone.
Secondary question being how to prevent the leaks, as several approaches are possible.

An example of a weak leak.
When VPPApiClient instance is created with async_thread [0] set to False, connect_internal does not create a thread [1] that would drain [2] message_queue, but each disconnect still puts an item [3] there.

My original motivation is suppressing the following error (not causing any failures in tests):
   Traceback (most recent call last):
     File "/usr/lib/python2.7/multiprocessing/queues.py", line 268, in _feed send(obj)
   IOError: [Errno 32] Broken pipe
but I have still not figured out how does that happen, except it is probably another kind of leak.

[0] https://github.com/FDio/vpp/blob/cbf1fd240318bf428f016c651be35f92e0bc13ca/src/vpp-api/python/vpp_papi/vpp_papi.py#L231
[1] https://github.com/FDio/vpp/blob/cbf1fd240318bf428f016c651be35f92e0bc13ca/src/vpp-api/python/vpp_papi/vpp_papi.py#L463-L467
[2] https://github.com/FDio/vpp/blob/cbf1fd240318bf428f016c651be35f92e0bc13ca/src/vpp-api/python/vpp_papi/vpp_papi.py#L699-L712
[3] https://github.com/FDio/vpp/blob/cbf1fd240318bf428f016c651be35f92e0bc13ca/src/vpp-api/python/vpp_papi/vpp_papi.py#L498