Random thought: the simplest way is probably to separate the visuals from the collisions. So, based on the example GIF, the visible bridge would probably not have a collision mesh at all, so it’s free to morph in any way it want. At the same time, there is an invisible collision mesh that can switch between two states (or switch between two entirely different collision meshes) depending on the state of the bridge (broken or complete).
In other words, when the visible bridge mesh is broken, there is an invisible collision mesh matching that configuration. When the user changes the bridge into its complete state, the collision mesh is swapped out and replaced with a different one. That’s pretty much how I would approach it at any rate.