[FEAT][ModelEvaluation] improve simple model comparison

There several ways in which this check could be improved:
1. Train and compare to several simple models options instead of just one
2. Added option of true random - randomly select a label from a uniform distribution of all labels (currently it is from true label distribution)
3. Think of more options for simple models
4. Add additional performance display options (maybe plot error graphs based on an interesting feature)
5. Select metrics / display based on task of model