Reassessing Layer Pruning in LLMs: New Insights and Methods