A proximal trust-region method for nonsmooth optimization with inexact function and gradient evaluations