From Uniform to Adaptive: General Skip-Block Mechanisms for Efficient PDE Neural Operators