You write chain = prompt | llm | parser. A teammate says this is identical to writing LLMChain(llm=llm, prompt=prompt) with a parser attached. What is the most important behavioral difference between LCEL pipe syntax and legacy LLMChain?