Datasets keeps rebuilding the dataset every time i call the python script

### Describe the bug

Every time it runs, somehow, samples increase.
This can cause a 12mb dataset to have other built versions of 400 mbs+

### Steps to reproduce the bug

`from datasets import load_dataset
s = load_dataset('~/.cache/huggingface/datasets/databricks___databricks-dolly-15k')['train']
`
1. A dataset needs to be available in the .cache folder
2. Run the code multiple times, and every time it runs, more versions are created

### Expected behavior

The number of samples increases every time the script runs

### Environment info

- `datasets` version: 3.6.0
- Platform: Windows-11-10.0.26100-SP0
- Python version: 3.13.3
- `huggingface_hub` version: 0.32.3
- PyArrow version: 20.0.0
- Pandas version: 2.2.3
- `fsspec` version: 2025.3.0