Linear Multi-Resource Allocation with Semi-Bandit Feedback