Allow for time-series variable weights

[bquistorff]

Another way to deal with lots of predictors is that, if they are from a set of time-series, is to have a constant weight per-time-series rather than per individual predictor. This is what MSCMT does (R library).