The current preference to promote 8bit images to floating point - possibly with dithering is very useful for JPGs and I am happy to have it as a default there -
promoting the high resolution JPGs to floating point with dither automatically does what I otherwise want to do for practically every JPG I open for editing, before my initial rotating and perhaps scaling down operations that further bring the histogram and content of the image to be proper floating point images.

For PNGs though - which also can be 8bit, the feature is a liability; by opening a PNG with intent to crop and export, one with the most basic workflow easily end up with a slightly dithered 16bit PNG, by limiting the feature to JPGs (and maybe some other mostly photographic 8bpc format?) workflows would be improved.