On the qualitative behaviour of numerical methods for delay differential equations of neutral type. A case study: Theta-methods