Since Python 2 reached its EOL Jan 1 2020 (over two years ago), I think that it's time to not actively improve Python 2 support any more.

Details:

1. We won't accept new PRs that improve Python 2 support, unless they fix a significant regression.
2. We won't drop or degrade existing support for Python 2, for now. Python 2 will become unsupported at some point, but at least not until the second half of 2022 (details TBD).
3. Existing open Python 2 PRs can still be merged if they seem useful and low risk.
4. We can remove some Python 2 specific test cases if they are hard to maintain (e.g. require runtime access to Python 2).

cc @sobolevn who has done some recent work on Python 2 features.