speeding up the predict function

[anyafries]

The predict function can be rather slow with many predictions.

For example, I think the following may speed up the function, but other improvements are welcome.

• add a functionality to the predict function to parallelize the main loop (which starts at line 201) e.g., using parallel::mclapply instead of the for loop and predict could have an extra argument cores.
• or try to reformat the code avoiding a loop entirely (maybe add extra dimension for num_preds and do as many matrix operations as possible)