Sort DataFrames inplace.

[evetion]

Fixes the performance of saving a Ribasim model in Python.

When we sorted, we assigned the returned dataframe to our model, triggering validation.

[Hofer-Julian]

Ruff claims:

Further, in many cases, inplace=True does not provide a performance benefit, as pandas will often copy DataFrames in the background.

Would be good to benchmark if that actually improve things.

[visr]

Can confirm that this no longer triggers validation on write.

[visr]

Turns out the Ubuntu CI failure was due to a missing custom sort method. Brought them all up to date with our docs in dc0c288. @evetion already mentioned that the sort algorithm on Linux could be slightly different, and we just luckily ended up with a sorted GeoPackage on Windows.

On top of that, we sort data when reading it into the core as well, so not sorting in Python should not lead to test failures. But those were similarly outdated, fixed in fe64269.