Make sure guess_density returns something with correct total charge

[antoine-levitt]

Unless I'm missing something, this is only approximately true currently with atomic pseudopotentials, and is not true at all with custom potentials