hiwe are using multi-lingual text embedding for similarity search and clustering.I see that once in a while, embedding models are being deprecated.my questions is: when moving to a newer embedding model, does it mean I need to re-embed my entire text...