Sharpe Ratio for skew-normal distributions: a skewness-dependent performance trade-off?