A Unifying Post-Processing Framework for Multi-Objective Learn-to-Defer Problems