Multi-Objective Recommendation via Multivariate Policy Learning