Plans context window allocation strategies when content exceeds model limits with compression options.