size_in_bytes the same for all splits

### Describe the bug

Hi, it looks like whenever you pull a dataset and get size_in_bytes, it returns the same size for all splits (and that size is the combined size of all splits). It seems like this shouldn't be the intended behavior since it is misleading. Here's an example:

```
>>> from datasets import load_dataset
>>> x = load_dataset("glue", "wnli")
Found cached dataset glue (/Users/breakend/.cache/huggingface/datasets/glue/wnli/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad)
100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 1097.70it/s]
>>> x["train"].size_in_bytes
186159
>>> x["validation"].size_in_bytes
186159
>>> x["test"].size_in_bytes
186159
>>>
```

### Steps to reproduce the bug

```
>>> from datasets import load_dataset
>>> x = load_dataset("glue", "wnli")
>>> x["train"].size_in_bytes
186159
>>> x["validation"].size_in_bytes
186159
>>> x["test"].size_in_bytes
186159
```

### Expected behavior

The expected behavior is that it should return the separate sizes for all splits.

### Environment info

- `datasets` version: 2.7.1
- Platform: macOS-12.5-arm64-arm-64bit
- Python version: 3.10.8
- PyArrow version: 10.0.1
- Pandas version: 1.5.2