On the Promise for Assurance of Differentiable Neurosymbolic Reasoning Paradigms