DEV: Scaled convergence criteria in NEB

Add-on to the NEB class. Convergence criteria can be scaled to focus optimization on saddle point and converge other interior states to a higher fmax. Convergence scaling is based on a displacement metric, where the fmax on each state depends upon the position of the state in relation to the highest point in the band. This can significantly reduce the number of force calls while giving identical results at the saddle point. Only implemented in series with dynamic_relaxation.