turns out not only sparc is borked, skip the checks on all 32 bit arches
with too large clongdouble alignments until we have an aligned
allocator.

mingw builds set the alignment requirement for complex doubles types to
16 byte so the tests checking the alignment flag must be disabled.

(debian) sparc system malloc does not provide the alignment required by 16 byte
long double types this means the inout intent cannot be satisfied and several
tests fail as the alignment flag can be randomly true or fals when numpy gains
an aligned allocator the tests could be enabled again.

BUG: #2408, Fix f2py Python 3 error message string bug.

BUG: #4256: f2py, PyString_FromStringAndSize is undefined in Python3.

The original was generating an exception message and, after aliasing,
calling PyBytes_AsString on a unicode string -> error. It was also
leaking references, although that probably didn't matter in context.

The fix here is on the cheap side, just use a C string for the message
without including the extra information about the erroneous type that
led to the exception.

No test, I don't know how to evoke this error.

Closes #2408.

Use PyUString_FromStringAndSize defined in npy_3kcompat instead. Not
using bytes may cause some problems, but strings seem like a better
choice. As modules generated with current f2py error out, this
particular use is not common and we are free to choose.

Closes #4256.

Replace "insinged_long_long" by "unsigned_long_long". Patch due to trac
user pepijndevos.

Closes #636.

ENH: add tobytes and stop using tostring in documentation

tostring returns bytes which are not equal to string, so provide a
tobytes function alias.
tostring does not emit a deprecation warning yet so rdepends do not need
to check two names to support older versions of numpy without warnings.