generally fix existing comparison plots

[nsfabina]

1. convert loss value and training time plots to bar plots

2. sort models by training times

3. make loss and validation plots with same scales

4. perhaps combine loss plots into single plot to make it clearer/easier to parse

5. show model names and not history paths

6. store weighted and unweighted loss in the model history object

7. differentiate between weighted and unweighted loss in the plots