Hello, I'm trying to integrate hybdrid search (embeddings are created with OpenAI embeddings), I've noticed when i input query without diacritics it has very poor performance in comparison when i enter query WITH diacritics. I haven't noticed such bad behavior directly in ChatGPt models. How is treated diacritics in TS when querying against embedding fields? I'm aware of that standard query is normalized (when se locale to blank or

en

) but what about query for embedding. I suppose that is taken as is and send againt embedding model in OpenAI? Thank you