consider switching from pytz to zoneinfo
Python 3.9 now has a zoneinfo package, backported up to 3.6+ in this pypi package.
According to this blog post from 2018, we should really switch to zoneinfo, particularly since we use it to load actual time zones in guess_zone
... (Actually, they suggest switching to dateutil.tz, but that advice is apparently outdated.)
That said, according to this LWN article from 2020, the pytz actually planned to make it a wrapper around the standard library, so we might actually be safe.
The task here, I guess, is to add a unit test that tries to reproduce the issues described in the bug report and, if present, fix the issue by switching away from pytz, or similar.
One issue with the switch is that the stdlib implementation is about a milisecond slower, which could be a problem in our quest for perforamnce (#25).