Trace Integrity for LLM Data Agents: A Vision for Auditable Structured Reasoning in Real-World Systems