jmoiron plays the blues

@jmoiron @dbenamy no, provided that it uses a timeout of 0 for items with no expiry (which is how JC's cache modification works)

@jmoiron @hintjens my issues were due to cruft left over after a fork; unrelated, can we have a diff error for s.connect('tcp://domainname:port')?

created gaspar, generic eventlet+zmq worker library

5789d354 to gaspar change the socket shutdown sequence to catch "Not connected error", which we don't care about, and fix the test suite on osx where eventlet does not build an epoll hub

f60ecfd7 to gaspar fix a weird issue where closed zmq sockets with linger set to 0 never get evicted from the eventlet hub; since this bug is not important to us and only shows up in unit testing, reset the hub after each test

756dc804 to gaspar adding a test that verifies zmq deals with big messages (the bug i was looking for ended up being in my own socket handling code)