Proprietary software doesn't have any problem with embedded code, for example TTF does exactly that with fonts and they're still being licensed.

Code embedding has been considered many times but the rabbit hole is way too deep. You are embedding unverified executable code, to begin with. Then there's diversity problem: too much code, too many implementation specifics with every quirk being able to mess up the result. All of that spanning multiple generations of color science pipelines, experts, hardware, and third-party software to interact with. Since the advent of ML (for denoising and segmentation in particular) you have to also embed the weights as well, making it outright impossible.

All this leads to RAW processing code going the way of scientific and financial code: once written, it's frozen and the substrate is emulated if necessary. Even the bugs are preserved.