dnn_superres upscaling support

[joseasoler]

OpenCV has implementations of specialized upscaling algorithms that todds currently does not support, which are described for example in here: https://towardsdatascience.com/deep-learning-based-super-resolution-with-opencv-4fd736678066. It should be possible to integrate these algorithms into the todds codebase and its pipeline.

Benchmarks and quality metrics: https://docs.opencv.org/4.x/dc/d69/tutorial_dnn_superres_benchmark.html

When upscaling is used, todds could use the original image as the first level mipmap, to minimize quality loss during mipmap calculations.