fix: remove START/END prints from run_baseline, convert to STEP format

inference.py

@@ -164,8 +164,7 @@ def run_llm_inference(client: OpenAI, MODEL_NAME: str, prompt: str) -> str:
 
 def run_baseline(env: CodeGuardianEnv) -> float:
     """Run a simple non-learning baseline agent for comparison."""
-    print("START: Running baseline agent...")
-    logger.info("[BASELINE START] Running baseline agent...")
+    print("STEP: Running baseline agent")
     
     obs = env.reset()
     total_reward = 0.0
@@ -191,8 +190,7 @@ def run_baseline(env: CodeGuardianEnv) -> float:
         
         logger.info(f"[BASELINE STEP] {step_count}: {step_type} -> reward={reward.value:+.3f}")
     
-    print(f"END: Baseline complete - Total reward: {total_reward:.3f}, Steps: {step_count}")
-    logger.info(f"[BASELINE END] Total reward: {total_reward:.3f}, Steps: {step_count}")
+    print(f"STEP: Baseline complete - total_reward={total_reward:.3f} steps={step_count}")
     return total_reward
 
 def update_agent_state(agent_state: AgentState, episode_reward: float, feedback: str) -> AgentState: