MongoClient.close() closes its sockets and signals its background threads to exit but it does not actually wait for the threads to exit (via join). This is problematic when using a MongoClient from a subinterpreter. Even when closing the MongoClient, an error gets raised when closing the subinterpreter: Py_EndInterpreter: not the last thread

Reported in: https://groups.google.com/d/msg/mongodb-user/vgsE6MpcyGE/1xm-WXMrAAAJ
and: https://groups.google.com/forum/#!topic/jep-project/lAv7-HVAnv8

Things to consider:

• Is the lack of a join() intentional?
• Don't hold the topology lock when attempting a join because the thread can be waiting for the same lock.
• PeriodicExecutor used to sleep for 100ms at a time to and check for a close but that was changed in PYTHON-983. Now PeriodicExecutors sleep for half a second at a time. Perhaps we should reintroduce the 100ms sleeps (without using condition variables on Python 2).
• Should we add a timeout parameter to MongoClient.close?
• Should we provide a MongoClient.join method with a timeout?