speed up compilation for functions with large numbers of temps (already shows for a couple of thousand, e.g. for large list literals)

This partially revert the change introduced in
767fce81.

Turns out AttributeNodes always start with type error, which we
don't want to propagate in some places.

Add visibility kwarg to cython.declare.

Add declarations for C++11/14 smart pointer factories (make_shared & make_unique)

DOCS: Fix typo

Don't error out when non-const value is passed as const function parameter

Silence a warning about applying unary minus to an unsigned integer

with MSVC.

function parameter.

parameter to a function. When function arguments are marked as const,
this means that the function should not modify them, not that
non-const arguments are not allowed.
This also fixes a test that began failing because this change resulted
in a slightly different error message when conversion to a const
qualified argument is not possible.

__Pyx_PyInt_TrueDivideObjC: switch comparison order

FROM_PY_FUNCTION: Cast to sdigit after unary minus, not before

The current order results in compiler warnings on 32 bit machines, e.g.

src/llfuse.c: In function '__Pyx_PyInt_TrueDivideObjC':
src/llfuse.c:43980:17: warning: left shift count >= width of type
                 if (8 * sizeof(long) <= 53 || (__Pyx_sst_abs(size) <= 52 / PyLong_SHIFT) || likely(labs(a) <= (1L << 53))) {

Switching the order so that the left shift is closer to the sizeof test
avoids the warning, presumably because it makes it easier for the
compiler to see that the left shift is only executed on 64 bit.

Thanks to Christian Neukirchen for doing most of the work!

The current order results in compiler warnings on 32 bit machines, e.g.

src/llfuse.c: In function '__Pyx_PyInt_TrueDivideObjC':
src/llfuse.c:43980:17: warning: left shift count >= width of type
                 if (8 * sizeof(long) <= 53 || (__Pyx_sst_abs(size) <= 52 / PyLong_SHIFT) || likely(labs(a) <= (1L << 53))) {

Switching the order so that the left shift is closer to the sizeof test
avoids the warning, presumably because it makes it easier for the
compiler to see that the left shift is only executed on 64 bit.

Thanks to Christian Neukirchen for doing most of the work!

The current cast isn't doing anything, because the unary minus implicitly
upcasts to int again, resulting in warnings about potentially value altering
conversions on 32 bit systems.

Fixes issue #877.