Please understand me correctly: Machine learning does have its use as an image editing. but not in raw development. Sure, VFX are fine, but the goal of raw development is to make the files that the camera has put out look as good as possible. And for that, machine learning is inadequate, because again, hallucinations and other defects. Once you have processed your image with the raw development software, sure. Machine learning, denoising, expansion and other VFX will certainly work. But my recommendation is to keep it out of raw development, not for purist reasons, but because there's genuinely no reason to use it, as we need to get precise results from raw first, and then we can add our VFX on top of that.

About the raw pipeline of Darktable, this is one of its greatest features. You can freely reconfigure it to suit your needs. By default, it uses a scene-referred workflow and you should really stick with that. But if you're an advanced user, you can freely shuffle the modules around as you like, just like you can do in DaVinci Resolve.

Edit: For beginners, really stick with the modules that you'll find in the different headers. Also, manipulate your modules starting from bottom to top, as this is the processing order for the modules.