HALO: Human Preference Aligned Offline Reward Learning for Robot Navigation