Try to adhere to scheduling deadlines when running callbacks

[skip ci]

This should result in the loop getting serviced more often when there
are many callbacks to run. This is hooked up to `getswitchinterval` to
allow control.

We check the interval every 50 callbacks. That's hardcoded, which I
don't like, but the 1000 callback number was hardcoded before, so it's
not really much of a change. A future extension would be to calculate
the check number dynamically based on dividing the switch interval.

I don't have a specific test case for this change because I'm a little
fried, but all the other tests pass. I did manually verify that the
example in #1071 matches expectations---that could probably be turned
into a test case with some work.

Fixes #1072
Fixes #1071
Fixes #1068

Make callbacks.c stop relying on implementation details of Cython and Python. Instead use documented ways to interact.

Revert the hack of copying the ares.h header distributed with c-ares to cares.h. Our ares.pyx's channel class doesn't need to be public.

Cleanup a mistake in test_httpservers.py, and make a change to the example portforwarder to better handle race conditions.

Try testing PyPy on Windows again. I'm getting clean-ish runs in my local test VM (with a small number of networking and subprocess tests disabled).

Run cythonize from setup.py

Add the pxd's as explicit deps; they don't actually seem to trigger recompilation, at least if just decorators are added.

cancel the whitespace test as part of the standard linting. there are too many false positives to keep up with in the generated C files. [skip appveyor]

This eliminates our sdist-time dependency on Makefile or make.cmd on
Windows. Fewer moving pieces are better.

Fixes #1076

Update corecext.ppyx

This eliminates the need to call cythonpp.py on it. It also leads to a
more consistent API.

We could take the same approach to bring back sig_fd if necessary, but
that's still not available to cython.

Attempt to workaround Visual C 9.0 for CPython 2.7 on windows missing stdint.h. Also correct a compiler warning.

We should be able to use dummy functions to link the child watchers on windows, and forbid them from being created. This gets us down to only one CPP def, LIBEV_EMBED

cython ignores the preprocessor directives we previously had in
libev.pxd, so that was doing nothing, and it was treating vfd_socket_t
just as something it had to turn into an int.

The function it used to do this was the same function it used to turn
a generic PyObject into an int, once all the typedefs were substituted
in (which does use the c preprocessor from libev_vfd.h). So it should
be functionally equivalent.

The vfd_* functions are no-op macros on POSIX, so there shouldn't be a
performance difference. The minor exception is that __dealloc__ is
always defined, and although it's a static function suitable to
inline, there is an extra Python API call to PyErr_Fetch and
PyErr_Restore. But if that makes a difference, I didn't spot it.