Heuristics for Combinatorial Optimization via Value-based Reinforcement Learning: A Unified Framework and Analysis