There exists one execution path that results in an IndexError, this fix solves it.

Specifically, whenever we evaluate a single scan which contains a match m that both has_gt and has_pred, in which all predictions are non-matched but also all of them are ignored and not counted as FP, then y_true and consequently y_true_sorted_cumsum are empty arrays.

Then accessing y_true_sorted_cumsum[-1] results in an IndexError.