Tune decision tree for Lee-Crippen SMARTS pKa calculation method

We have freshly implemented Lee-Crippen SMARTS pKa calculation method. Using new datasets from various sources we need to adjust the decision SMARTS-tree, which is an essential part of this method, to produce good pKa values for various compound with different structure.

At the beginning the following datasets can be used:
[Machine learning meets pKa](https://github.com/czodrowskilab/Machine-learning-meets-pKa/tree/master/datasets
)
[DataWarrior pKa in water](https://github.com/TVect/mol_property/blob/master/mol_property/pka/train/data/pKaInWater.csv)
[IUPAC dataset](https://github.com/IUPAC/Dissociation-Constants)