Increase number of items for which list.extend([...]) is optimised. Benchmarking shows that append() is still faster for 8 items (on an empty list) than extend().
This limit could be increased for larger lists (which grow by a larger amount), but if we have to settle on one value then 8 is good enough, as we can expect literal item lists to be rather short.

Keep "result_is_used" flag across node replacements in type analysis and optimisations to make sure it's still available at code generation time.

Fixes #2059.

This causes the compiler to get confused and reuse the magic `cimport cython` module. See #2057.

Closes #2048.

Undeprecate __richcmp__

Silence C compiler warning about fall-through in switch statement for kw-only argument parsing code.

Tweak method call fallbacks of dict.get(), dict.setdefault() and dict.pop() into fast unbound method calls to the underlying C function.

Fix #2047: optimize dict.pop() using _PyDict_Pop()

This reverts commit e1ae63dd.

Streamline __Pyx_PySet_Remove() helper function. Add comments about unclean return values of new helper functions.
Closes #2042.

Issue #2042: add optimized versions of `x in set`, set.remove() and set.discard()