Chained Information-Theoretic bounds and Tight Regret Rate for Linear Bandit Problems