Implement an Agent Final Review Diagnostic

Implement diagnose_agent(config, trace, eval_results) that produces a structured diagnostic of an agent system. config is a dict with keys like {"loop_type": str, "tools": list[str], "memory_enabled": bool, "guardrails_enabled": bool, "max_steps": int}. trace is a list of step dicts, each with: - step (int): iteration number. - tool (str or null): tool called this step. - status (str): "success", "failed", "blocked", or "timeout". - latency_ms (int): step duration. - tokens (int): tokens used. eval_results is a dict with {"score": float, "categories": {"correctness": float, "safety": float, "efficiency": float, "robustness": float}}. Requirements: - Classify failures by dimension: loop, tools, memory, rag, safety, reliability, cost. - Check config for missing components (no guardrails, no memory, low max_steps). - Analyse trace for patterns: repeated tool calls (looping), high failure rate, timeouts, excessive token usage per step. - Check eval_results for low-scoring categories. - Prioritise root causes: missing components > critical failures in trace > low eval scores. - Recommend remediations for each issue found. - Return {"overall_rating": str, "issues": list[dict], "priorities": list[str], "review_plan": list[str]}. - Each issue is {"dimension": str, "severity": "critical"|"major"|"minor", "finding": str, "remediation": str}.