Maintain row order after cross join

[AdrianSosic]

Fixes the failing polars tests.

By default, polars gives no guarantees on the resulting row order of a join (see here), meaning that our tests used to pass just by luck. This has changed since polars==0.19.0, which apparently included changes that affect the row order of our test dataframes. The PR fixes these tests by ignoring the row order during the equality check.

The current version of the polars cross join computing the Cartesian product currently makes no guarantees on the resulting row order (see here). While strictly not a bug, this makes the behavior inconsistent with the corresponding pandas implementation and makes comparison more difficult. Accordingly, our tests started to fail since polars 0.19.0 which apparently included changes that affect the row order of our test dataframes.

Enforcing the pandas-equivalent row order seems like the best option for now, while potentially not exploiting the maximum possible speed of the polars join. In case speed really becomes a limitation, we can remove the restriction but then need to check that we don't rely on the order anywhere.

[Scienfitz]

see comment about performance
if performance is impacted by the fix I prefer not to include it and soften tests instead