As I see it the gain would be a more consistent and intuitive API. (less surprises)
The cost is one word size increase for the listener. Like I perceive the desing goals of ChibiOS/RT the change would be worth the cost, but this is your decision to make.
About breaking compatibility: Code that correctly clears the flags before
waiting on a new event would still work.
And I doubt that anyone relies on the fact that a flag creates a new event when not cleared.