Rethinking explainability: toward a postphenomenology of black-box artificial intelligence in medicine