Add CAST expression support to parser

[nssalian]

Closes #198

Picks up where #209 left off with @Fokko's feedback addressed. Thanks @jayceslesar for the original work.

Rationale for this change

Adds CAST(column AS type) parsing that maps to Iceberg transforms (date → DayTransform, year → YearTransform, month → MonthTransform, hour → HourTransform). Also implements BoundTransform.ref() and eval() — both were missing abstract methods required by BoundTerm. Introduces UnboundTransform with bind() that validates via can_transform().

Are these changes tested?

Yes. 11 new tests covering all transform types, case insensitivity, nested fields, unsupported types, comparison operators, and boolean composition.

Are there any user-facing changes?

No.

[nssalian]

@Fokko @geruh @kevinjqliu PTAL

[Fokko]

For testing, I think it would be good to have an integration test as well. The StringLiteral should be converted to a DateLiteral when bound to the transform.

[nssalian]

Added a unit test for this.

[rambleraptor]

I've been playing around with this PR and I ran into something.

Here's a quick demo script for your own testing.

schema = Schema(NestedField(1, "ts", TimestampType(), required=False))

sql = "CAST(ts AS year) > 2020"
row = Record(datetime_to_micros(datetime(2024, 1, 1)))
print(expression_evaluator(schema, parser.parse(sql), case_sensitive=True)(row)) # False

It looks like we're running into this issue because we coerce ts to a TimestampType at comparison time. That means the right side is TimestampLiteral(2020) -> which isn't the year 2020.

[rambleraptor]

We should have more tests like this with uneven types. I posted an example in another comment.