In most basic pathfinding algorithms, edges have no maximum size. In practice what we usually do is to filter the actual graph when feeding it to the pathfinding algorithm to filter out channels that are smaller than the amount being routed. The basic pathfinding algorithms are unaware of capacities. Similarly, we can just make the capacity of the split out channels equal to the total capacity of the mechanism.
This may be different for flow-based pathfinding algorithms as you proposed for renepay before. But I suspect the flow-based pathfinding would need deeper changes to consider that each mechanism has more variables now than just x and 1 - x.