Sure is a neat demonstration of their stability and control, but I can't think of single reason for a robot to do a handstand or somersault.
The reason is exactly stated in the opening of your statement. Having such capabilities will become important when faced with the challenges that will be encountered in our natural world.
AI, as I see it, is more important in the choice of actions to be taken where simple rules - such as stability - are not as straightforward.
I agree about the opening part of the statement. But hand stands are a very, very fringe part of the real world. I have successfully made it through several decades of life with the only application for a handstand being in a high school gym class. I am sure the vast majority of those on this forum find handstands similarly relevant to their lives. A somersault is a useful maneuver for entertainment or as recovery from a fall (at least for humans). Robots probably are a long ways from needing entertainment, and may have very different optimal responses to accidents. Just as Atlas' method for resuming a standing position after a fall only vaguely resembles common human methods.
Boston Dynamics has always done well on real world examples from their earliest four legged devices on snow and ice up through Atlas navigating piles of broken concrete blocks. They have the three sigma cases of maneuvering around the world well in hand. The problem remains how to make their devices autonomously do useful things. Their demo of stacking (carefully bar coded) boxes on shelves was a good but very limited example. I don't know if that is the limit of their success, or if it is only the limit of what they are able and willing to publicly reveal.