Generics

[delfick]

This PR is the result of nearly 10 months of me tackling this a bit at a time every few weeks.

All I wanted to do was this:

    it "can make generic types":
        container = reg.create(Container[ItemOne], {"category": "one"})
        assert isinstance(container.item, ItemOne)
        assert container.item.one == 20
        assert container.item.two == 50
        assert not container.item.three

        container2 = reg.create(Container[ItemTwo], {"category": "two"})
        assert isinstance(container2.item, ItemTwo)
        assert container2.item.one == 3
        assert container2.item.two == 5
        assert container2.item.four

Turns out python type annotations aren't very nice to introspect and you have to do some fiddly things to understand when a type is optional or annotated or generic and how inheritance changes things.

I also wanted to make it so registering creators could be more powerful, which resulted in the need to be able to sort type annotations by complexity so that I could deterministically and efficiently find the most appropriate creator for a given type. With this sorting capability I'm able to look at the most specific creators before the most generic creators.

I also took a rabbit hole into resolving python typing ForwardRefs. This is where a type annotation is represented as a string at runtime because it was for a type not available at the time it was referenced. If these aren't resolved then strcs can't find the correct creator for that field. I will rejoice when I can make strcs support 3.13 and remove this resolving logic because of https://peps.python.org/pep-0649/

I've also slipped in package upgrades to use hatchling.