Risk-aware inverse design of char yield and energy quality in solid waste pyrolysis using test-validated machine learning surrogates