Lets the user choose the maximum number of iterations the robust
estimator runs for, similary to findHomography. This can significantly
improve performance (at a computational cost).