bug(?): Polars / pandas row order in pivot differs

[MarcoGorelli]

Noticed in test_pivot_no_index from #1640, the tests fail if I remove .sort("ix", "bar")

gonna have to check this

[MarcoGorelli]

(Pdb) df.to_native()
shape: (4, 4)
┌─────┬─────┬─────┬─────┐
│ ix  ┆ col ┆ foo ┆ bar │
│ --- ┆ --- ┆ --- ┆ --- │
│ i64 ┆ str ┆ i64 ┆ str │
╞═════╪═════╪═════╪═════╡
│ 1   ┆ a   ┆ 1   ┆ x   │
│ 1   ┆ b   ┆ 2   ┆ y   │
│ 2   ┆ a   ┆ 3   ┆ z   │
│ 2   ┆ b   ┆ 4   ┆ w   │
└─────┴─────┴─────┴─────┘
(Pdb) df.to_native().pivot('col', values='foo')
shape: (4, 4)
┌─────┬─────┬──────┬──────┐
│ ix  ┆ bar ┆ a    ┆ b    │
│ --- ┆ --- ┆ ---  ┆ ---  │
│ i64 ┆ str ┆ i64  ┆ i64  │
╞═════╪═════╪══════╪══════╡
│ 1   ┆ x   ┆ 1    ┆ null │
│ 1   ┆ y   ┆ null ┆ 2    │
│ 2   ┆ z   ┆ 3    ┆ null │
│ 2   ┆ w   ┆ null ┆ 4    │
└─────┴─────┴──────┴──────┘

🤔 is Polars respecting maintain_order here? The grouped columns are 'ix' and 'bar', yet the output isn't sorted according to them

[MarcoGorelli]

I think this wasn't quite behaving as-documented in Polars: pola-rs/polars#20403

I'm not sure we should be guaranteeing an output order at all here in Narwhals, and we should remove maintain_order