test-pytorch-model: add a model that matches tflite outputs

add argmax, max, and box enconding transform layers so that
we can match TFLite outputs:
* detection_boxes
* detection_classes
* detection_scores
* number_of_detections