Add option for named splits when using ds.train_test_split

### Feature Request 🚀

Can we add a way to name your splits when using the `.train_test_split` function?

In almost every use case I've come across, I have a `train` and a `test` split in my `DatasetDict`, and I want to create a `validation` split. Therefore, its kinda useless to get a `test` split back from `train_test_split`, as it'll just overwrite my real `test` split that I intended to keep.

### Workaround

this is my hack for dealin with this, for now :slightly_smiling_face:

```python
from datasets import load_dataset
​
​
ds = load_dataset('imdb')
ds['train'], ds['validation'] = ds['train'].train_test_split(.1).values()
```