The RAIL Principles for Neurosymbolic AI: Reasoning, Assurances, Interfacing and Learning