Current saga state at the time of compensation.
This includes all state updates made by steps that executed after this step (up to the point of failure). This enables "smart compensation" where earlier steps can make decisions based on later execution progress.