@gadolf Why bump such an old thread? Do you have questions yourself?
Not sure if this answers your collective questions: ML is constantly being updated. It has a nightly base with all of the possible features but not all cameras support them. Moreover, each hardware and firmware difference between cameras has to be considered. There are lots of reverse engineering involved and investigation of hardware registers. It is painstaking work even if you have the best and brightest and then you would have to connect to code that also needs to be imagined and developed.
The ETTR + Dual ISO module combo, in tandem with other features (edit e.g., enfuse stacking), make ML very compelling for DR. Sure, it can be awkward to manage because it is not a part of the original package, so you would need to adapt your capturing and post-processing to ML. But if you have the programming chops and the latest camera supported by ML, with enough RAM and power, you could script a lot of the stuff to make post-processing easier afterward. Just consider the heat budget before you try using the camera innards as a supercomputer. 