Generalized mean p-values for combining dependent tests: comparison of generalized central limit theorem and robust risk analysis