Fix bug in detection_map_op.

[qingqing01]

The Output(AccumPosCount) is a LoDTensor with a shape [class_number, 1], but following code will get the shape: [max_class_id + 1, 1], the max_class_id + 1 is not alwasy equal to class_number.

https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/fluid/operators/detection_map_op.h#L186

    int max_class_id = 0;
    int true_pos_count = 0;
    int false_pos_count = 0;
    for (auto it = label_pos_count.begin(); it != label_pos_count.end(); ++it) {
      int label = it->first;
      if (label > max_class_id) max_class_id = label;
      int label_num_pos = it->second;
      if (label_num_pos == 0 || true_pos.find(label) == true_pos.end())
        continue;
      auto label_true_pos = true_pos.find(label)->second;
      auto label_false_pos = false_pos.find(label)->second;
      true_pos_count += label_true_pos.size();
      false_pos_count += label_false_pos.size();
    }

    int* pos_count_data = output_pos_count.mutable_data<int>(
        framework::make_ddim({max_class_id + 1, 1}), ctx.GetPlace());

The Output(AccumTruePos) and Output(AccumFalsePos) also have problem.