Fixes #222.

Rather than provide a messy monkey-patch on top of `fcntl`, it's simpler
just to mention this attribute, which seems to be stable, and has at
least one reported use of working.

[skip ci]

This time, I saw a bunch of errors like this, which maybe suggest something is
leaking between tests, but I don't know what:

  ===================================================
  ERROR: test_quit (__main__.Test)
  ----------------------------------------------------------------------
  Traceback (most recent call last):
    File "C:\projects\gevent\src\greentest\greentest.py", line 289, in wrap_restore_handle_error
      return method(self, *args, **kwargs)
    File "C:\projects\gevent\src\greentest\greentest.py", line 278, in wrap_error_fatal
      return method(self, *args, **kwargs)
    File "C:\projects\gevent\src\greentest\greentest.py", line 161, in wrap_timeout
      return method(self, *args, **kwargs)
    File "test__backdoor.py", line 75, in test_quit
      conn = self._create_connection()
    File "test__backdoor.py", line 46, in _create_connection
      conn.connect(('127.0.0.1', self._server.server_port))
    File "C:\Python34-x64\lib\site-packages\gevent\_socket3.py", line 291, in connect
      r = getaddrinfo(address[0], address[1], self.family)
    File "C:\Python34-x64\lib\site-packages\gevent\_socketcommon.py", line 272, in getaddrinfo
      return get_hub().resolver.getaddrinfo(host, port, family, socktype, proto, flags)
    File "C:\Python34-x64\lib\site-packages\gevent\resolver_thread.py", line 65, in getaddrinfo
      return self.pool.apply(_socket.getaddrinfo, args, kwargs)
    File "C:\Python34-x64\lib\site-packages\gevent\pool.py", line 326, in apply
      return self.spawn(func, *args, **kwds).get()
    File "C:\Python34-x64\lib\site-packages\gevent\event.py", line 382, in get
      self._wait_core(timeout, ())
    File "C:\Python34-x64\lib\site-packages\gevent\event.py", line 106, in _wait_core
      result = self.hub.switch()
    File "C:\projects\gevent\src\greentest\greentest.py", line 468, in switch
      return _original_Hub.switch(self, *args)
    File "C:\Python34-x64\lib\site-packages\gevent\hub.py", line 630, in switch
      return RawGreenlet.switch(self)
  OSError: [Errno 9] File descriptor was closed in another greenlet

  ----------------------------------------------------------------------
  Ran 6 tests in 10.319s

Hopefully this will help diagnose the timeout error seen (only) on
Appveyor: The test that runs after test_multi hangs; all the other tests
work. (This was verified by changing the name of test_multi and thus its
sort order; the test after it is the one that fails.)

Refactor test__backdoor.py to be simpler, and add a test case for this
specific issue. Take the proposed patch of switch in by overriding
throw() just for backdoor because I don't really see a general place for
it in the hub.

It suddenly started crashing with an overflow getting the MRO of our
class that extends build_ext. No versions have changed other than cython
though, so I don't know why it crashed.

Log here: https://travis-ci.org/gevent/gevent/jobs/170750778

Simplify subprocess.py: Unify many py2/py3 branches

This adds the *timeout* and *input* arguments for Py2.

This required a small doctest renormalizer to handle the byte string
difference.

Fix test__makefile_ref under Windows 10 with new psutil 4.4 which actually started returning open files, not just connections.

Fix test__makefile_ref under Windows 10 with new psutil 4.4 which actually started returning open files, not just connections.

Also temporarily disable a bunch of travis/appveyor builds, I'm
expecting a bunch of commits on this branch.

Support for PyPy3

One more test that fails on Travis. Although we're getting lots of failures there now as it works to deal with the backlog of builds due to the DNS issues (we're timing sensitive).

These seem to be a mix of Py3.3 and newer tests.

Make them all "pass" by disabling as appropriate; almost all of them
were already disabled for either Python 2.7.8 (old SSL implementation)
or Python 3.4 (signal handling). I didn't see anything to be concerned
about.

Fixes #826.

Fix test__semaphore.py under PyPy3 by raising the right (wrong, really) exception type that PyPy does.