From Equations to Insights: Unraveling Symbolic Structures in PDEs with LLMs