Clean up and repair the type spec generation for the limited API, leaving more of the work to the TypeSlots instead of special-casing all sorts of things.

Ideally, most of the code that is uninteresting for users should be out of the way and not reside before the translated user code.
Mark all code section name beginnings in the C code file to make them easier to follow and move around.

Fix C-code indentation of PyModuleDef struct by avoiding duplicate opening braces in conditional code.

* Changed always_allow_keywords_T295 for Py3.9

https://bugs.python.org/issue37645 appears to change some error
messages slightly (including the qualname and not just the
function name). This change just allows doctest to cope with
both the new and old messages

This reverts commit 5204d869.
See https://github.com/cython/cython/issues/3548

Work around error that "__reduce_cython__ cannot be found" when re-importing an extension module that was already initialised before.
Backport to 0.29.x.
Closes #3545.

* Work around error that "__reduce_cython__ cannot be found" when re-importing an extension module that was already initialised before.
* Exclude "reimport_failure" test in Py2 and PyPy since Py2 does not allow reimports and PyPy3 apparently does nothing for them.

Closes #3545.

* Updated URL for the "embedding Python" link in external_C_code.rst
  in the userguide.
* The freenet.org.nz domain has clearly expired at some point and now
  the original link goes spam links, but there are copies in the
  Wayback Machine.
* The updated (fixed) URL uses the most recent archive for the
  original page (author: June, 2004; web archive: Feb, 2012).

generate_type_definitions maintained a list of ctuples for which code
was already generated, but the logic did not match the logic used in
generate_type_header_code, which does not generate code if the entry is
marked as unused.

If the first ctuple of a certain type was marked unused, but later
instances were not, this could result in no code being generated for
the ctuple at all. Resolve by verifying whether the ctuple entry is used
before adding it to the list.
Co-authored-by: Kirk Meyer <kirk.meyer@ravenind.com>

generate_type_definitions maintained a list of ctuples for which code
was already generated, but the logic did not match the logic used in
generate_type_header_code, which does not generate code if the entry is
marked as unused.

If the first ctuple of a certain type was marked unused, but later
instances were not, this could result in no code being generated for
the ctuple at all. Resolve by verifying whether the ctuple entry is used
before adding it to the list.
Co-authored-by: Kirk Meyer <kirk.meyer@ravenind.com>

* Updated URL for the "embedding Python" link in external_C_code.rst
  in the userguide.
* The freenet.org.nz domain has clearly expired at some point and now
  the original link goes spam links, but there are copies in the
  Wayback Machine.
* The updated (fixed) URL uses the most recent archive for the
  original page (author: June, 2004; web archive: Feb, 2012).

Follows Python behaviour, but excludes "__pyx_…" names in utility code.

Closes GH-1382.

Currently excludes PyPy2.
Closes GH-2781.

Add "tp_pypy_flags" field to PyTypeObject struct when compiling in PyPy 6.0+ in order to avoid C compiler warnings about an uninitialised struct field.

Right now it's used in a few simple cases (function call and single assignment)
Don't attempt to remove reference types.
Be wary of the "safety check" for double moves when things like result() are called twice
Made sure fallback option was genuinely c++03 complient so that test should pass on clang

This reverts commit 52635828.

multiprocessing requires some additional setup on Windows, which I don't currently want to implement in setup.py.