Assessing the Properties of Patient-Specific Treatment Effect Estimates from Causal Forest Algorithms Under Essential Heterogeneity