AnandaSky: A Vision-Language Model for Line-Level Transcription of Historical Sinographic Documents