use at least ssize_t temp variables for buffer index calculations even if users use smaller integer types as we might be doing wrap-around calculations that depend on the potentially larger buffer shape

Rewrite __Pyx_sst_abs using only preprocessor macros, and add a case for 64-bit MSVC

Comparing `sizeof(Py_ssize_t)` to different sizes at compile time results in warnings or errors when the C++ compiler tries to find a function matching `abs(value)` (Per the standard library spec, no functions match if `ssize_t` is a typedef of `long long`). Luckily, `pyport.h` provides some SIZEOF macros we can use at preprocess time. __Pyx_sst_abs is now conditionally defined to a much shorter expression.

64-bit MSVC is one case where `ssize_t` is a `long long`. Instead of falling back to `((value<0) ? -value : value))`, I added a check for 64-bit MSVC that calls the MS-specific function `_abs64(long long)`

This change also includes the change given by PR #402, improved with a check against the available C++ version as `std::abs` is only available on C++11.

fix binary search in code object cache

Enable overloading the assignment operator.

Conflicts:
	setup.py

Conflicts:
	CHANGES.rst
	Cython/Shadow.py

rename "gi_running" attribute of coroutine type to "cr_running" and adapt some docstrings to the type split

(start + end) can overflow.

re-implement Coroutine.__await__() using a dedicated wrapper type to avoid having to make a Coroutine half an Iterator