For machine-learning of potential energy surfaces, we need the electronic energy and gradient, but many electronic structure theory packages return only the total energy (including nuclear-nuclear repulsion) and its gradient with respect to nuclear positions.
We should support reading the nuclear-nuclear repulsion energy where it is possible (e.g. Gaussian log file) and otherwise, add a field to the IOData object with the electronic energy and its gradient
For consistency with other fields in IOData, I'd propose the names
energy_el
atgradient_el