How do you handle context window limitations and token cost optimization when processing extremely long, text-heavy financial spreadsheets at Reducto?
