Originally, as part of WT-5677, we started work for rollback_to_stable.  That specific work is going into this ticket.  Basically it involves 1) having the timestamp thread advance the stable point 2) in each thread, keeping the state as of the stable point when it is passed, 3) periodically letting all worker threads finish, then call rollback_to_stable, and checking that the current state matches the saved state.  One preliminary piece was some refactoring to test/format to make this feasible.