Enable parametric DGP with new covariates in each replication

First figure out, whether this is a valid use-case. Currently, I think there are papers where the evaluation is designed in such a way that covariates are sampled anew for each replication. 
However, our `generate_data` method in `justcause.data.utils` is not capable of handling such parametric DGPs. 

