QM40 Dataset for large molecule QM property prediction

[elilaird]

This PR adds the newly released QM40 dataset from the paper [1]. This dataset follows a similar structure to the QM9 dataset, though with a different feature order (details in docstring).

QM40 is a QMx type of dataset which includes 150K molecules optimized from B3LYP/6-31G(2df,p) level of theory in the Gaussian16 with QM parameters, optimized coordinates, Mulliken charges and Local vibrational mode parameters as a quantitative measurer of the bond strengths. These 150,000 molecules have been chosen to represent the real chemical space of drug-like compounds. The molecules have a maximum heavy atom count of up to 40 and can contain the following atoms: Carbon (C), Fluorine (F), Oxygen (O), Nitrogen (N), Sulfur (S), and Chlorine (Cl).

[1] Madushanka, A., Moura, R.T. & Kraka, E. QM40, Realistic Quantum Mechanical Dataset for Machine Learning in Molecular Science. Sci Data 11, 1376 (2024). https://doi.org/10.1038/s41597-024-04206-y