Review OpenCV code
Our opencv.cpp file contains functions that transform image data, but assume that images are in 3-channel RGB. We very often have monochrome images, so their processing takes much longer than it should, with RRGGBB reordering to BGRBGR, operation, channel splitting and memory copy.
Image data is often duplicated in our OpenCV functions, while OpenCV matrices support external data buffers, as used in mpp in the three last functions of this file: https://gitlab.com/free-astro/siril/blob/mpp/src/opencv/opencv.cpp .
Reviewing both memory allocation and the handling of monochrome images will improve performances a lot in operations like rotation, scaling, homography transformation, deconvolution, backgroundnoise (because of FinestScale). I suggest we do this for 0.9.11 since it's in the performance theme and not very long to do.