feat(environment): call reward_computer.reset() in reset() to clear novelty scorer

server/environment.py

@@ -39,6 +39,9 @@ class RedTeamEnvironment:
         self.is_active      = True
         self.created_at     = datetime.utcnow()
 
+        if self.reward_computer and hasattr(self.reward_computer, "reset"):
+            self.reward_computer.reset()
+
         logger.info(f"Episode started: {self.episode_id}")
 
         return RedTeamObservation(