Computing Minimally Unsatisfiable Subformulas: State of the Art and Future Directions