BUG: move weakref-able implementation to base C extension type (fix PyPy)

[jorisvandenbossche]

Closes #1575

I haven't yet checked if this actually fixes it on PyPy, but it's a potential fix for the issue. My understanding is that on PyPy also extension types defined in C support weakref by default, so therefore adding this manually in the subclasses as we did here is failing (https://foss.heptapod.net/pypy/pypy/-/issues/2670#note_45293)

The implementation of adding a weakreflist slot is mimicked from numpy (how they do that for the ndarray extension type)

[coveralls]

Pull Request Test Coverage Report for Build 3266840634

• 1 of 1 (100.0%) changed or added relevant line in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage remained the same at 84.631%

---

Totals
Change from base Build 3265218526:	0.0%
Covered Lines:	2153
Relevant Lines:	2544

---

💛 - Coveralls

[EwoutH]

Out of curiosity, did you figure out why PyPy 3.9 didn't work in the CI but 3.8 did?

[jorisvandenbossche]

Out of curiosity, did you figure out why PyPy 3.9 didn't work in the CI but 3.8 did?

Sorry, missed that question. I am not really sure anymore why I switched from 3.9 to 3.8 for PyPy, but so the reason the CI passed in the end is because the commit after that to skip some tests that don't work on PyPy, not because of the 3.9->3.8 switch.