Issue with train_test_split maintaining the same underlying PyArrow Table

### Describe the bug

I've been using the train_test_split method in the datasets module to split my HuggingFace Dataset into separate training, validation, and testing subsets. However, I've noticed an issue where the split datasets appear to maintain the same underlying PyArrow Table.

### Steps to reproduce the bug

1. Load any dataset ```dataset = load_dataset("lhoestq/demo1")```
2. Try the next code:
```python
from datasets import Dataset, DatasetDict

train_size = 0.6

split_train = dataset["train"].train_test_split(
train_size=train_size,
)

separate_dataset_dict = DatasetDict({
"train": split_train["train"],
"test": split_train["test"],
})
```
3. The next code ```print(separate_dataset_dict)``` when printing the dataset it gives the indication that they have 3 and 2 rows respectively.
4. But the next code:
```python
print(len(separate_dataset_dict["train"].data['id']))
print(len(separate_dataset_dict["test"].data['id']))
```

Indicates that both tables still have 5 rows.

### Expected behavior

However, I've noticed that train_test_split["train"].data, test_val_split["train"].data, and test_val_split["test"].data are identical, suggesting that they all point to the same underlying PyArrow Table. This means that the split datasets are not independent, as I expected.

I believe this is a bug in the train_test_split implementation, as I would expect this function to return datasets with separate underlying PyArrow Tables. Could you please help me understand if this is expected behavior, or if there's a workaround to create truly independent split datasets?

I would appreciate any assistance with this issue. Thank you.

### Environment info

I tried in Colab:

- `datasets` version: 2.13.0
- Platform: Windows-10-10.0.22621-SP0
- Python version: 3.10.11
- Huggingface_hub version: 0.14.1
- PyArrow version: 12.0.0
- Pandas version: 2.0.1

and my PC:

- `datasets` version: 2.13.0
- Platform: Linux-5.15.107+-x86_64-with-glibc2.31
- Python version: 3.10.12
- Huggingface_hub version: 0.15.1
- PyArrow version: 9.0.0
- Pandas version: 1.5.3