Fix fit_to_window Python embed_preprocessing()

[Wovchena]

Close #99

The implementation was incorrect. It performed scales based interpolation instead if sizes. Padding was missing. Now it is aligned with C++.

[sungchul2]

Thanks for quick fix.
Although I didn't check, some functions use metadata related to fit_to_window and used into postprocess might be affected, such as

model_api/model_api/python/openvino/model_api/models/detection_model.py

Lines 136 to 146 in c828050

	def resize_detections_with_aspect_ratio(
	detections, original_image_size, resized_image_size, model_input_size
	):
	scale_x = model_input_size[0] / resized_image_size[0] * original_image_size[0]
	scale_y = model_input_size[1] / resized_image_size[1] * original_image_size[1]
	for detection in detections:
	detection.xmin *= scale_x
	detection.xmax *= scale_x
	detection.ymin *= scale_y
	detection.ymax *= scale_y
	return detections

What do you think?

[Wovchena]

Updated