Improving TTS for Shanghainese: Addressing Tone Sandhi via Word Segmentation