Breaking the √T Barrier: Instance-Independent Logarithmic Regret in Stochastic Contextual Linear Bandits