Support multiple speeds in predictions

@mlerik multiply tot_dist with speed/inverse of speed:

            # Register possible conflict
            if self.predictor and num_steps < self.max_prediction_depth:
                int_position = coordinate_to_position(self.env.width, [position])
                if tot_dist < self.max_prediction_depth:
                    pre_step = max(0, tot_dist - 1)
                    post_step = min(self.max_prediction_depth - 1, tot_dist + 1)

                    # Look for opposing paths at distance num_step
                    if int_position in np.delete(self.predicted_pos[tot_dist], handle):
                        conflicting_agent = np.where(np.delete(self.predicted_pos[tot_dist], handle) == int_position)
                        for ca in conflicting_agent:
                            if direction != self.predicted_dir[tot_dist][ca[0]]:
                                potential_conflict = 1
                    # Look for opposing paths at distance num_step-1
                    elif int_position in np.delete(self.predicted_pos[pre_step], handle):
                        conflicting_agent = np.where(self.predicted_pos[pre_step] == int_position)
                        for ca in conflicting_agent:
                            if direction != self.predicted_dir[pre_step][ca[0]]:
                                potential_conflict = 1
                    # Look for opposing paths at distance num_step+1
                    elif int_position in np.delete(self.predicted_pos[post_step], handle):
                        conflicting_agent = np.where(np.delete(self.predicted_pos[post_step], handle) == int_position)
                        for ca in conflicting_agent:
                            if direction != self.predicted_dir[post_step][ca[0]]:
                                potential_conflict = 1