Text this: Combining Autoregressive Models and Phonological Knowledge Bases for Improved Accuracy in Korean Grapheme-to-Phoneme Conversion