Add dataframe safeguard

Created by: AMHermansen

This PR aims to introduce a safeguard to _predict_as_dataframe which should prevent runtime errors, which are results of #570 (closed).

The safe guard, currently just either crops or appends to the "prediction_columns" list. This is perhaps too crude, but hopefully a user would be able to fill in the correct variables in the dataframe/csv-file later.

Do you have any suggestions or other approaches to deal with this @RasmusOrsoe

requested review from @jprado

mentioned in issue #570 (closed)

closed