Handling of large integer literals is wrong when combining integers of different types/dynamic integers

[itamarst]

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 pytest

import polars as pl
 
from tests.unit.conftest import INTEGER_DTYPES
 
@pytest.mark.parametrize("int_dtype", INTEGER_DTYPES)
def test_literal_integer_addition(int_dtype: pl.DataType) -> None:
    one = pl.lit(1, dtype=int_dtype)
    for i in range(100):
        value = 2**i
        assert pl.select(pl.lit(value, dtype=pl.Int128) + one).item() == value + 1
        assert pl.select(pl.lit(-value, dtype=pl.Int128) + one).item() == -value + 1
        assert pl.select(pl.lit(value) + one).item() == value + 1
        assert pl.select(pl.lit(-value) + one).item() == -value + 1

Log output

The test fails, because select(...).item() often turns into None.

Issue description

This is a continuation of the issue I discovered when diagnosing #20807.

If you look through the logic in polars-core/src/utils/supertype.rs, it hasn't been updated for Int128. One part is covered in #20830, but this is a broader issue.

Things that need fixing include (but are likely not limited to):

• materialize_smallest_dyn_int()
• materialize_dyn_int_pos()
• most significantly, their caller get_supertype_with_options()

None of them acknowledge the existence of Int128 dtype, and therefore give unexpected results for large integers.

Expected behavior

See the assertions in the test.

Installed versions

Polars git as of Jan 20, 2024.