Make h2.events more typing-friendly

[mhils]

This PR adds proper constructors for most of the h2 events. This allows us to tighten the type annotations, simplifying type checks in downstream implementations. These events should not be instantiated by anyone except h2 itself, so it shouldn't be a breaking change.

Based on #1297.

[Kriechi]

LGTM! Thanks!

These events should not be instantiated by anyone except h2 itself, so it shouldn't be a breaking change.

I assume you intended this as "most users will not use this API and won't be affected by this change"?
And not in the way of "this API should be made private and documented as do-not-use"?

Please add a quick note into the changelog. Thanks!

[mhils]

I assume you intended this as "most users will not use this API and won't be affected by this change"?

If by "this API" you mean the __init__ method, then yes.

I will add constructors for the other events and then also a changelog entry.

[mhils]

I've switched to @dataclass to make things simpler and less repetitive. What do you think?

Python 3.9 requires some extra logic because it doesn't have kw_only yet. We can drop kw_only entirely if you'd like, even though I think it's generally a good idea.

[Kriechi]

I do like dataclasses!

I do like the kw_only pattern!

I don't like the extra fuzz about py3.9. But since py3.9 becomes EOL later this year, I think its acceptable. Happy to hear other thoughts on it!

[mhils]

This is ready for review now. With these changes we get from 31 mypy complaints down to 3 in mitmproxy (see linked PR for the fixes). All other downstream consumers should profit similarly.

[Kriechi]

Please add a changelog entry.

The docs should be auto-generated, or do you think we need to update something?
https://github.com/python-hyper/h2/blob/master/docs/source/api.rst#events

LGTM! Thanks!