I am aware that producing JPEG files results in lossy compression, and that there is a lossless JPEG format, but I actually wonder if the standard JPEG format can be "hacked" in a certain way that allows it to store uncompressed pixels.
One seemingly possible way I have found is decreasing the MCU size from 8x8 to 1x1. If the format itself allowed such a change, would the color information remain unchanged for the whole image, or would the compression algorithm still mess with it?
No matter what you do you are going to get some attempt at compression and some loss in "standard" jpeg. The would be the case even if you set all your quantization tables to {1, 1, ....} and did not subsample.
You are going to get "loss" during the RGB => YCbCr transformation. You get going to get additional loss in the DCT.