New HF-Net models with NMS

hfnet_vino_480x640_nms1r1: hfnet_vino model integrated non-maximum suppression (NMS) with radius=1 and iterations=1.

HF-Net models for OpenVINO and TensorFlow

The TensorFlow model can take any image size (can be specifed in hfnet_tf.py).
The OpenVINO models take images with a specific size.
With image inputs of the same size, the OpenVINO model should be significantly faster than the TensorFlow model.
Both run on CPU.