Theoretical characterisation of the Gauss-Newton conditioning in Neural Networks