FORCE_REDOWNLOAD raises "Directory not empty" exception on second run

### Describe the bug

Running `load_dataset(..., download_mode=datasets.DownloadMode.FORCE_REDOWNLOAD)` twice raises a `Directory not empty` exception on the second run.

### Steps to reproduce the bug

I cannot test this on datasets v2.11.0 due to #5711, but this happens in v2.10.1.

1. Set up a script `my_dataset.py` to generate and load an offline dataset.
2. Load it with
```python
ds = datasets.load_dataset(path=/path/to/my_dataset.py,
name='toy',
data_dir=/path/to/my_dataset.py,
cache_dir=cache_dir,
download_mode=datasets.DownloadMode.FORCE_REDOWNLOAD,
)
```
It loads fine
```
Dataset my_dataset downloaded and prepared to /path/to/cache/toy-..e05e/1.0.0/...5b4c. Subsequent calls will reuse this data.
```
3. Try to load it again with the same snippet and the splits are generated, but at the end of the loading process it raises the error
```
2023-04-11 12:10:19,965: DEBUG: open file: /path/to/cache/toy-..e05e/1.0.0/...5b4c.incomplete/dataset_info.json
Traceback (most recent call last):
File "", line 2, in
File "/path/to/conda/environment/lib/python3.10/site-packages/datasets/load.py", line 1782, in load_dataset
builder_instance.download_and_prepare(
File "/path/to/conda/environment/lib/python3.10/site-packages/datasets/builder.py", line 852, in download_and_prepare
with incomplete_dir(self._output_dir) as tmp_output_dir:
File "/path/to/conda/environment/lib/python3.10/contextlib.py", line 142, in __exit__
next(self.gen)
File "/path/to/conda/environment/lib/python3.10/site-packages/datasets/builder.py", line 826, in incomplete_dir
shutil.rmtree(dirname)
File "/path/to/conda/environment/lib/python3.10/shutil.py", line 730, in rmtree
onerror(os.rmdir, path, sys.exc_info())
File "/path/to/conda/environment/lib/python3.10/shutil.py", line 728, in rmtree
os.rmdir(path)
OSError: [Errno 39] Directory not empty: '/path/to/cache/toy-..e05e/1.0.0/...5b4c'
```

### Expected behavior

Regenerate the dataset from scratch and reload it.

### Environment info

- `datasets` version: 2.10.1
- Platform: Linux-4.18.0-483.el8.x86_64-x86_64-with-glibc2.28
- Python version: 3.10.8
- PyArrow version: 11.0.0
- Pandas version: 1.5.2