Occasional Travis failure under 3.4; cannot reproduce locally and the code is a duplicate of stdlib. [skip ci]

Account for different default bufsize under py3; avoid a resourcewarning under Py3 with GEVENT_FILE=thread; the incorrect bytes return in universal newlines should be fixed.

Fix subprocess under Py3 when using FileObjectThread. Fix catching the right exception on downlevel python versions.

Fix subprocess doctests under py3; fix raising the right exception if a child fails to exec under py2

Import the 3.4 version of test_subprocess and make it (mostly) pass. Includes a test for #374; fixes #374.

Add support for the new onerror attribute in CFFI 1.2.0

Add support for the new onerror attribute in CFFI 1.2.0/PyPy 2.7.0. This fixes our signal handling issues.

Support more objects that implement the buffer protocal by asking for the len() of their memoryview, not the len() of themselves.

Also potentially reduces the number of calls to memoryview so may be slightly more efficient.

Fixes #466.

lint; those tests were fixed with the previous commit but travis is having issues getting builds off the ground

If the environment had not been monkeypatched when gevent's ssl module
was imported, they could get the stdlib socket, meaning that reads and
writes could block greenlet switching for an arbitrary amount of time.
This was especially visible on SSL servers.

Fixes #597.

It preserves the tracebacks and obsoletes the ``apply_e`` function.

Add test cases for this behaviour.

Fixes #556.

This ensures ThreadPool has the same API, including accepting multiple iterables in imap.

Also, capture the _raise_exception method of a failed greenlet and use it to raise the original traceback from those methods.