Fix: convert lists, ndarrays to tuples when sorting unit test dataframes

[georgesittas]

This was an oversight which probably surfaced with a delay, hopefully this fixes the CI in main.

See comment below.

[georgesittas]

DuckDB recently bumped to 0.10.1 which included a commit that changed the dataframe list representation to be np.ndarray instead of python's list, so the instance check in _to_hashable broke.

The workaround here was to check for both lists and np.ndarrays and convert them to tuples. The other case where non-hashables are stringified was implemented in anticipation of dictionary support in unit tests.