[FEA] Support grouped list aggregation and explode in cuDF-Polars

Grouped list aggregation is not currently supported by the GPU engine, which also blocks workflows that aggregate values into lists and then explode them back into rows.

```python
def test_grouped_list_aggregation_explode(engine: GPUEngine) -> None:
lf = pl.LazyFrame(
{
"g": [1, 1, 2, 2, 2],
"x": [10, 20, 30, 40, 50],
}
)

q = lf.group_by("g").agg(pl.col("x")).explode("x")

assert_gpu_result_equal(q, engine=engine, check_row_order=False)
```

Related broad inventory issue: #23151

Start with the grouped list aggregation and explode behavior described in test_grouped_list_aggregation_explode, then trace the cuDF-Polars GPU engine's grouped aggregation entry point. Use the shown LazyFrame query as the regression case and compare its result with the CPU engine; done means the GPU result passes assert_gpu_result_equal with row-order checking disabled. Related inventory issue: #23151.

Written by the indexing model from the issue text.