Add a `st.timezones()` strategy using the stdlib `zoneinfo` module

[Zac-HD]

Python 3.9 adds the PEP 615 zoneinfo module to the standard library, and IMO we should add a corresponding hypothesis.strategies.timezones() - with #2414 to skew the distribution and help it find bugs.

I'm open to either adding a dependency on the backport (zoneinfo ; python_version<"3.9"), or just raising a runtime error if you try to use the strategy when zoneinfo is not available; leaning towards no dependency to avoid imposing any cost on people who aren't using it.

[Zac-HD]

Note that @pganssle wrote the tricky bit, a valid_keys() strategy for zoneinfo keys like UTC or Australia/Sydney, for python/cpython#22863. Ideally we merge our version before CPython merges that, so they can use the new strategy from the start. Paul: I assume you can offer that strategy under the MPL-2.0 for Hypothesis?

I'm therefore leaning towards adding two strategies: st.timezone_names() for the names, and a convenience st.timezones() strategy implemented as return timezone_keys().map(zoneinfo.ZoneInfo) (similarly, the dateutil extra will be refactored as .map(dateutil.tz.gettz) to share #2414 heuristics).

[Zac-HD]

Prototype: master...Zac-HD:zoneinfo

(needs tests and review by @pganssle, but I think it should mostly work)