This handles None exception values that would otherwise fail when passed
to traceback.print_exception() and prevent the process from exiting. An associated test is also included.

Fix test__subprocess for Py3. A premature close of half the pipe was causing the whole pipe to get closed, apparently, at least under OS X, so don't close it. The test still fails if the fix is removed.

I'm not 100% sure this is the correct fix; could there be some underlying issue? The subprocess code is the same on both Py2 and Py3.

pywsgi mostly works, there are a few timeout issues lurking somewhere in the test suite. But it's enough for the example servers to run.

Fix several more socket-related tests under Py3: test__examples, test__os, test__fileobject, test__subprocess_poll, test__server, test__execmodules.

First pass on socket._fileobject for Py3; this went away, but is used heavily in our code base, so the shortest path is probably going to be to emulate it. This fixes test__backdoor.py. others to come.

One test in test_threading.py fails under PyPy 2.6; that test has been removed from upstream (https://bitbucket.org/pypy/pypy/diff/lib-python/2.7/test/test_threading.py?diff2=4ecd94a66a87&at=vendor/stdlib) so don't run it here either. I'm not updating the whole module yet because we run tests under 2.5.0 on Travis.

subprocess.Popen `bufsize` option actually used

Use the TravisCI container infrastructure and enable linters

Previous debugging shows the socket accepting all the data we send it, which is just under 2MB. Maybe travis containers have very large default socket buffers? Up the data size to about 4MB.

Pre-allocate the (biggec) data before we start taking timing measurements, and be sure to close the socket when done.