Calculate error on observable calculation based on number of samples

We currently take the S(Q,w) which has been calculated from the simulation trajectory as the ground truth. However, as has been shown by @MBartkowiakSTFC there is an intrinsic large variability at low energies (since to get an accurate value at zero energy would require an infinite length of simulation) in the calculated S(Q,w) based on nominally identical simulations, which differ only by their initial seeds.

We should include a standard deviation of the calculated S(Q,w) points, so that our forcefield optimisation (and data scaling) concentrates on the parts of the simulation which we are most certain about.