predictions for network data that does not include all cohorts

[mriech]

If the function runGompitzPrediction is applied to a subset of data that does not cover all cohorts included in the calibration object, stratum names are allocated falsely in returned object.

There are several ways of fixing this:

• delete relevant stratum from calibration object
• Include fake data for all cohorts in network data and delete after prediction
• just correct allocation of stratum names, maybe via manipulating the stratum matrix