bug: rendering transcripts with an Assistant terminal prompt

[bderenzi]

If we're using basic routing and have a terminal prompt and all the "worker" prompts and the terminal prompt are OpenAI assistants, then the transcript double renders the response from the worker prompt used and the terminal prompt. (It may not matter whether the work prompts are assistants vs base LLM -- I didn't test)

More info:
https://dimagi.slack.com/archives/C05PK63Q89H/p1727473171673049