Payoff-based learning with matrix multiplicative weights in quantum games