Text this: Self-Prioritizing Multi-Agent Reinforcement Learning for Conflict Resolution in Air Traffic Control with Limited Instructions