Can you validate your reasoning?
Personally I found the opposite by not using state machines and drive the locomotion via blend spaces and blend by nodes.
With out writing a book I found using blend spaces and blend nodes produces much faster inputs, cleaner blends, and a lot more data driven compatible.
Excluding the need for a movement component to make things easier the question is who should own the animation blueprint used to drive the animations?
Personally I’m holding off until plug-ins is added to the marketplace, or a root motion component, as be it in-place or root motion they are still both just another data set.
Or
At the very least Epic adds a RM sample. 