Contextualized Speech Recognition: Rethinking Second-Pass Rescoring with Generative Large Language Models | Synapse