Predicates with & where one side is is_in(Series) are no longer pushed down to pyarrow scan

[nameexhaustion]

Checks

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of Polars.

Reproducible example

import polars as pl
import pyarrow.dataset as ds

path = ".env/data.parquet"
pl.DataFrame({"x": 1}).write_parquet(path)
lf = pl.scan_pyarrow_dataset(ds.dataset(path))

print(lf.filter(pl.col("x").is_in([2])).explain())
# PYTHON SCAN []
# PROJECT */1 COLUMNS
# SELECTION: col("x").is_in([Series])
print(8 * "=")
print(lf.filter((pl.col("x") == 1) | (pl.col("x").is_in([2]))).explain())
# FILTER [([(col("x")) == (1)]) | (col("x").is_in([Series]))] FROM
#   PYTHON SCAN []
#   PROJECT */1 COLUMNS

Log output

No response

Issue description

This is due to a combination of https://github.com/pola-rs/polars/pull/17939/files#diff-e1f991b2a947bfb3c1d98696b1256362ee233eef35b8a86715e4eabceeb46b02 and the expression not being considered streamable

ref #17045 (comment)

Expected behavior

The 2nd query should also push the predicate into a SELECTION at scan

Installed versions

main / 1.4.1

[ritchie46]

Ah, right. That's due to IO source. I can take this one.