Compatibility of expected utility and μ/σ approaches to risk for a class of non location–scale distributions