predict_pls() has a bug when data is matrix

If data is specified as matrix, the function errors:
```
Parallel encountered this ERROR: 
8 nodes produced errors; first error: incorrect number of dimensionsError in model$construct_scores[as.character(c(1:nrow(model$data))), ] : 
  subscript out of bounds
```

I think the problem is subsetting is different for matrices and dataframes