The option was removed on the win32 platform regardless of actual
compiler, leading to an error.

These fixes are not agressive as some of the code is complicated
and it is better to be careful.

The files numeric.py and numerictypes.py are not easily analysed
and the latter is self modifying. Pyflakes generates a number of
invalid warnings for those files.

Update version hash in numpy/core/code_generators/cversions.txt.
Update API version in numpy/core/setup_common.py
Update API version in numpy/core/include/numpy/numpyconfig.h

The library functions casin, casinh, catan, catanh are inaccurate
in glibc version < 2.16, so use the numpy fallback functions for
those.

Set the HAVE_XXXX macros for all of the following functions.

    csin,   ccos,   ctan
    casin,  cacos,  catan
    csinh,  ccosh,  ctanh
    casinh, cacosh, catanh
    cexp,   clog,   cpow
    cabs,   csqrt,  carg
    cimag,  creal,  conj,
    cproj

The macros do not need to be used, but it is good to have all of
them available.

More efficient as it uses processor overflow flags if available.

When compiler uses link time optimization it will not emit the binary
representation of the long double during linking. This prevents building
numpy with -flto.
To avoid this try linking the object when the compile detection failed.
With this change numpy can be built with link time optimization with 4 worker
processes using following command:

CC='gcc -fno-fat-lto-objects -flto=4 -fuse-linker-plugin -O3' \
LDSHARED='gcc -fno-fat-lto-objects -flto=4 -fuse-linker-plugin -shared
-O3' AR=gcc-ar \
python setup.py build_ext

See also:
https://www.mail-archive.com/numpy-discussion@scipy.org/msg45535.html

Should also work with icc's -ipo switch (probably closes gh-4992)

wraps the C99 cube root function cbrt.