Bayesian Learning in Sparse Graphical Factor Models via Variational Mean-Field Annealing