Fix the track 1 with the intensity and add the train/dev/test split

[AvataGarde]

1. Now for the track 1, it could follow the Quality Control examples to take the average intensity for each emotion and then compare with the threshold(0.5) to decide the final label. It is also possible for each instance that has multiple emotions.
2. Because Neutral has no intensity, if for a text if it has no emotions, then in track 1 task, it would be assigned 1 on the 'Neutral ' emotion.
3. Add the function to split the train/dev/test set, and it would allocate at least 5 rarest instances to the dev set