Fix constructor signatures and test annotations for ty

[rambleraptor]

Rationale for this change'

ty exposes out 3,547 type errors. This gets us down to 655 type errors. These errors all stem from the fact that mypy is ignoring errors relating to optional dependencies.

These come from two different situations:

1. We need __init__ methods on Pydantic subclasses. This ends up being an issue on expressions mostly. I found a GitHub issue on Pydantic that talks about the same issue. The duplicative __init__ methods are annoying, but I don't see any way around it.
2. There's a bunch of tests where we have to explicitly pass in client=None since it won't be coming from an untyped dictionary. This actually feels more correct, since client isn't a property. This is only occurring in tests and it's a semantic difference. This feels like splitting hairs with the type checker, but I'm happy to let it do its thing.

Are these changes tested?

Run uv run ty check.

Are there any user-facing changes?

[kevinjqliu]

should we make ty part of make lint? I dont think #3964 does this

[kevinjqliu]

LGTM

2 nits

[rambleraptor]

@kevinjqliu we very much should not...yet. Mypy and ty have conflicting views on which # type: ignore statements are necessary / not. Some of those type: ignore statements are necessary because mypy believes the pyarrow import resolves as an Any... which isn't true and the source of our issues here.

Once we get ty in a good place, the last PR will have to be removing the unnecessary type: ignore statements as we move over to ty.