Python.h header to fix a compilation error with OpenMP. PyThreadState_GET()
becomes an alias to PyThreadState_Get() to avoid ABI incompatibilies.

It is important that the _PyThreadState_Current variable is always accessed
with the same implementation of pyatomic.h. Use the PyThreadState_Get()
function so extension modules will all reuse the same implementation.

Test self.theclass, not datetime. Regression introduced by manual tests.

of datetime.datetime: microseconds are now rounded to nearest with ties going
to nearest even integer (ROUND_HALF_EVEN), instead of being rounding towards
zero (ROUND_DOWN). It's important that these methods use the same rounding
mode than datetime.timedelta to keep the property:

   (datetime(1970,1,1) + timedelta(seconds=t)) == datetime.utcfromtimestamp(t)

It also the rounding mode used by round(float) for example.

Add more unit tests on the rounding mode in test_datetime.

_PyTime_Divide() rounding was wrong: copy code from Python default which has
now much better unit tests.

Oops, I forgot to document my change.

PyObject_Length() returns a P_ssize_t, not an int. Use a Py_ssize_t to avoid
overflow.

On Windows, the tv_sec field of the timeval structure has the type C long,
whereas it has the type C time_t on all other platforms. A C long has a size of
32 bits (signed inter, 1 bit for the sign, 31 bits for the value) which is not
enough to store an Epoch timestamp after the year 2038.

Add the _PyTime_AsTimevalTime_t() function written for datetime.datetime.now():
convert a _PyTime_t timestamp to a (secs, us) tuple where secs type is time_t.
It allows to support dates after the year 2038 on Windows.

Enhance also _PyTime_AsTimeval_impl() to detect overflow on the number of
seconds when rounding the number of microseconds.

On Windows, the tv_sec field of the timeval structure has the type C long,
whereas it has the type C time_t on all other platforms. A C long has a size of
32 bits (signed inter, 1 bit for the sign, 31 bits for the value) which is not
enough to store an Epoch timestamp after the year 2038.

Add the _PyTime_AsTimevalTime_t() function written for datetime.datetime.now():
convert a _PyTime_t timestamp to a (secs, us) tuple where secs type is time_t.
It allows to support dates after the year 2038 on Windows.

Enhance also _PyTime_AsTimeval_impl() to detect overflow on the number of
seconds when rounding the number of microseconds.

* test_eintr: support verbose mode, don't redirect eintr_tester output into
  a pipe
* eintr_tester: replace os.fork() with subprocess to have a cleaner child
  process (ex: don't inherit setitimer())
* eintr_tester: kill the process if the unit test fails
* test_open/test_os_open(): write support.PIPE_MAX_SIZE bytes instead of
  support.PIPE_MAX_SIZE*3 bytes

Skip test_open() and test_os_open(): both tests uses a FIFO and signals, but
there is a bug in the FreeBSD kernel which blocks the test. Skip the tests
until the bug is fixed in FreeBSD kernel.

Remove also debug traces from test_eintr:

* stop using faulthandler to have a timeout
* remove print()

Write also open and close on two lines in test_open() and test_os_open()
tests. If these tests block again, we can know if the test is stuck at open or
close.

test_eintr: don't always run the test in debug mode.

import_init() imports the "_imp" module, not the "imp" module.

format_stack(), and extract_stack() called without arguments.

format_stack(), and extract_stack() called without arguments.

format_stack(), and extract_stack() called without arguments.

Fix test_write(): copy support.PIPE_MAX_SIZE bytes, not support.PIPE_MAX_SIZE*3
bytes.

sys.stderr is sometimes a StringIO. The redirection was just a hack to see
eintr_tester.py output in red in the buildbot output.

ProactorEventLoop now supports SSL.

Some test_eintr hangs on waiting for the child process completion if an error
occurred on the parent. Kill the child process on error (in the parent)
to avoid the hang.