Modelling retrial-upon-conflict systems with product-form stochastic Petri nets