Stream requires replay when adding use LLM block and not changing schema

When adding a use LLM block to my existing stream, it requires a replay. I cannot replay because it outputs a backing dataset with a lot of processed data that I do not want to reprocess. My schema did not changed. How can I avoid this replay and deploy the changes only on new data?

Hey I don’t think this is supported today but tagging in @svercillo who can chime in from the streaming team