Simple Primitives With Feasibility- and Contextuality-Dependence for Open-World Compositional Zero-Shot Learning