Adding "h" argument to calculate future observations.

[dfalbel]

@bearloga here is an initial patch addressing #4.

2 points I am still in doubt:

• Naming: I think it would be more intuitive to call the new variables as y_1, y_2, y_3 instead of mlts_lead_1. It also helps when using dplyr-like functionality like select(dt, starts_with("mlts")) for features and select(dt, starts_with("y")) for response variables.
• Dealing with missing values Like lagging, we loose 1 observation for each leading variable added. We already filter out the missing observations for lagged variables. Should we do the same for when leading?