Generating train split takes a long time

### Describe the bug

Loading a simple webdataset takes ~45 minutes.

### Steps to reproduce the bug

```
from datasets import load_dataset
dataset = load_dataset("PixArt-alpha/SAM-LLaVA-Captions10M")
```

### Expected behavior

The dataset should load immediately as it does when loaded through a normal indexed WebDataset loader. Generating splits should be optional and there should be a message showing how to disable it.

### Environment info

- `datasets` version: 2.20.0
- Platform: Linux-4.18.0-372.32.1.el8_6.x86_64-x86_64-with-glibc2.28
- Python version: 3.10.14
- `huggingface_hub` version: 0.24.1
- PyArrow version: 16.1.0
- Pandas version: 2.2.2
- `fsspec` version: 2024.5.0