[FEA] Support broadcasting aggregated expressions in groupby/rolling for cudf-polars

```python
import polars as pl

pl.Config.set_verbose(True)

lf = pl.LazyFrame(
{
"a": [1, 2, 3, 4, 5, 6],
"b": [1, 2, 3, 4, 5, 6],
"g": ["x", "x", "x", "y", "y", "y"],
}
)

print(
lf.group_by("g")
.agg((pl.col("a") - pl.col("a").mean()).dot(pl.col("b") - pl.col("b").mean()))
.collect(engine="gpu")
)

"""
/opt/miniconda3_py312/lib/python3.12/site-packages/polars/lazyframe/frame.py:2407: PerformanceWarning: Query execution with GPU not possible: unsupported operations.
The errors were:
- NotImplementedError: Broadcasting aggregated expressions in groupby/rolling
return wrap_df(ldf.collect(engine, callback))
"""
```

The polars version is '25.12.00a362'