Inferring sub-/super-levels may introduce mistakes

[rmnldwg]

The following example illustrates the issue with the function infer_and_combine_levels():

>>> from lydata import load_datasets
>>> clb = next(load_datasets(year=2021, institution="clb"))
>>> clb.iloc[16].pathology.ipsi
Ia       NaN
Ib     False
II     False
III     True
IV     False
V        NaN
VII      NaN
Name: 16, dtype: object
>>> infer_and_combine_levels(clb).iloc[16].pathology.ipsi
Ia     False
Ib     False
II     False
III     True
IV     False
V        NaN
VII      NaN
I      False
IIa    False
IIb    False
Va      None
Vb      None
Name: 16, dtype: object

Note how the involvement in LNL Ia was suddenly changed to False, although it is unknown.