SW-SpeedDLM: Sliding Window Speculative Decoding for Diffusion Language Models Under Long Context Constraints | Synapse