Reinforcement Learning Based Temporal Logic Control with Soft Constraints Using Limit-deterministic Generalized Buchi Automata