Closing the Computational-Query Depth Gap in Parallel Stochastic Convex Optimization