Maximum principle for delayed stochastic mean-field control problem with state constraint