Combinatorial Bandits with Linear Constraints: Beyond Knapsacks and Fairness