[Data] supper passing pyarrow.dataset.Expressions to Dataset.filter's expr

[schmidt-ai]

Description

Currently, one may only pass a string representation of an Expression to Dataset.map(expr...), but it would be nice to allow passing either a string or an Expression.

Use case

I want to programmatically generate a filter expression:

from functools import reduce
from operator import or_

cols = ["a", "b"]

# filter rows where any of `cols` is null
expr = ~reduce(or_, [pa.field(col).is_null(nan_is_null=True) for col in cols])

dataset.filter(expr=expr)

and I'd rather do this with the native pyarrow Expression API instead of string manipulation.

If I try dataset.fiilter(expr=str(expr) it raises ValueError: Invalid syntax in the expression.