Discuss and potentially change use of aK from DsgeSmoother

The smoothed forecasts (filtered_K_step_ahead variables) are written into a matrix aK with size [max(options_.filter_step_ahead) nvars nobs+max(options_.filter_step_ahead)] That is, instead of writing the 1:K step ahead prediction error made for time 1 into aK(1,:,1) we have a triangular structure where they are in aK(1:K,:,1) to aK(1+K,:,1+K) While changing the arrangement in aK seems too much trouble, we might want to change the way aK is written into oo_.FilteredVariablesKStepAhead as it is currently counterintuitive. If one has 100 observations and wants 8 step ahead forecasts, oo_.FilteredVariablesKStepAhead is 108 instead of 100 columns long. I think we should have all forecasts made at time t arranged in basically column t instead of distributed over columns t to t+K.