I have stumbled upon an edge case where Gemini Flash 2 would use the same tool call ids on multiple consecutive AI messages within a thread (not within a parallel tool call). The post_model_hook ...