feat(typing): make types more precise for many common Values

[NickCrews]

This partially mitigates #11680, at least for some of the more common types.

In particular, it solves this ask:
#11680 (comment)

I focused on the most common use cases, eg StringValue + "foo" is more important than NumericColumn.kurtosis()

This fixes several sorts of problems:

• It types methods as returning the right datashape. Sometimes this was as simple as use Self, but often this required @overload because the method depends on the datashape of second args, or it is an eg string->int method
• It make datatypes more precise in some places, eg before it was simply returning NumericValue, but we actually know it is FloatingScalar
• It fixes the typing in some places. eg in StringValue.__getitem__, it actually accepts IntegerColumn as well, not just IntegerScalar, as it was typed before. Or IntegerValue.convert_base() incorrectly said it returned a IntegerValue, but it actually returns a StringValue

[cpcloud]

Unless there's something I'm missing here I don't see how return-type-only overloads are actually useful.

[cpcloud]

I don't think this guaranteed to be an IntegerColumn. Consider the example in the docstring.

[NickCrews]

Thanks you're right

[NickCrews]

Adjusted to ir.FloatingColumn

[cpcloud]

Why do we need overloads for these? You can't use overload for return values anyway. Consider this example:

@overload
def foo(x: int) -> str: ...
@overload
def foo(x: int) -> int: ...

y = foo(1)  # which overload is called?

There's no straightforward way to compute the type of y.

[NickCrews]

What I was going for was more of

@overload
def foo(x: SubtypeOfInt) -> str: ...
@overload
def foo(x: int) -> int: ...

so that the first overload would be selected as the first matching case. I can do some more tests though to ensure that this is actually happening.

[cpcloud]

@NickCrews Can't you split the changes that make return types more precise out into a separate PR that doesn't contain the changes with all the additional overloads?

[NickCrews]

Yes!

[NickCrews]

OK in that new version I removed all overloads, but I did go through even more methods and made them more correct and accepting of plain python values and Deferreds, eg adjusting where:ir.BooleanValue | None = None to where:bool | ir.BooleanValue | Deferred | None = None

[cpcloud]

LGTM, thanks!