feat(polars): add ArrayUnion operation

[IndexSeek]

Description of changes

Implements ArrayUnion with pl.Series.list.set_union.

Reasoning for test marker adjustment

I was also running into the NaN issue that occurs with DataFusion. Here's some code I was using to test this.

In [1]: import polars as pl
   ...: from ibis.interactive import *

In [2]: left, right = [[3, 2], [], []], [[1, 3], [None], [5]]

In [3]: df = pl.DataFrame([left, right], schema=["left", "right"])

In [4]: df
Out[4]: 
shape: (3, 2)
┌───────────┬───────────┐
│ left      ┆ right     │
│ ---       ┆ ---       │
│ list[i64] ┆ list[i64] │
╞═══════════╪═══════════╡
│ [3, 2]    ┆ [1, 3]    │
│ []        ┆ [null]    │
│ []        ┆ [5]       │
└───────────┴───────────┘

In [5]: duck_con = ibis.connect("duckdb://")
   ...: polars_con = ibis.connect("polars://")

In [6]: for con in (duck_con, polars_con):
   ...:     t = con.create_table("t", df, overwrite=True)
   ...:     expr = t.left.union(t.right)
   ...:     print(t.get_backend().name)
   ...:     print(con.execute(expr))
   ...: 
duckdb
0    [1, 2, 3]
1       [None]
2          [5]
Name: ArrayUnion(left, right), dtype: object
polars
0    [3.0, 2.0, 1.0]
1              [nan]
2              [5.0]
Name: ArrayUnion(left, right), dtype: object

The following is mentioned in the docs at https://docs.pola.rs/user-guide/expressions/missing-data/#notanumber-or-nan-values:

In pandas by default a NaN value in an integer column causes the column to be cast to float. This does not happen in Polars - instead an exception is raised.

I wonder if this happens when going to pandas by way of execute.