Stochastic Subgradient Descent Escapes Active Strict Saddles on Weakly Convex Functions