WGRAMMAR: Leverage Prior Knowledge to Accelerate Structured Decoding