A benchmark for time-series reasoning and retrieval based on five distinct long context dataset