Log Agent Loop Steps to a Trace

Implement log_step(trace, step_data) that appends a step to an agent execution trace and returns the updated summary. trace is a list of step dicts (may be empty for the first step). step_data is a dict with keys: - step_number (int, required) - thought (str, required) - action (str, required — the tool name or "Final Answer") - arguments (dict, optional — defaults to {}) - observation (str, optional — defaults to "") - duration_ms (int, optional — defaults to 0) Return a summary dict: { "total_steps": len(trace) + 1, "last_step": step_number, "last_action": action, "tool_calls": <count of steps where action is not "Final Answer">, "total_duration_ms": <sum of all duration_ms across trace + step_data>, "thoughts": <list of all thought strings in order> } Edge cases: - If step_data is None or missing required keys (step_number, thought, action), return None without modifying the trace. - If trace is None, treat it as an empty list. - Compute total_steps as the current number of steps after appending.