Inconsistent performance of the "arabic_billion_words" dataset

## Describe the bug
When downloaded from macine 1 the dataset is downloaded and parsed correctly.
When downloaded from machine two (which has a different cache directory),
the following script:

import datasets
from datasets import load_dataset
raw_dataset_elkhair_1 = load_dataset('arabic_billion_words', 'Alittihad', split="train",download_mode='force_redownload')

gives the following error:

**Downloading and preparing dataset arabic_billion_words/Alittihad (download: 332.13 MiB, generated: 1.49 GiB, post-processed: Unknown size, total: 1.82 GiB) to /root/.cache/huggingface/datasets/arabic_billion_words/Alittihad/1.1.0/687a1f963284c8a766558661375ea8f7ab3fa3633f8cd9c9f42a53ebe83bfe17...
Downloading: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 348M/348M [00:24<00:00, 14.0MB/s]
Traceback (most recent call last):
File ".../why_mismatch.py", line 3, in
File "/opt/conda/lib/python3.8/site-packages/datasets/load.py", line 1632, in load_dataset
builder_instance.download_and_prepare(
File "/opt/conda/lib/python3.8/site-packages/datasets/builder.py", line 607, in download_and_prepare
self._download_and_prepare(
File "/opt/conda/lib/python3.8/site-packages/datasets/builder.py", line 709, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/opt/conda/lib/python3.8/site-packages/datasets/utils/info_utils.py", line 74, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
datasets.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='train', num_bytes=1601790302, num_examples=349342, dataset_name='arabic_billion_words'), 'recorded': SplitInfo(name='train', num_bytes=0, num_examples=0, dataset_name='arabic_billion_words')}]**

Note that the package versions of datasets (1.15.1) and rarfile (4.0) are identical.

## Steps to reproduce the bug
import datasets
from datasets import load_dataset
raw_dataset_elkhair_1 = load_dataset('arabic_billion_words', 'Alittihad', split="train",download_mode='force_redownload')

# Sample code to reproduce the bug

## Expected results
Downloading and preparing dataset arabic_billion_words/Alittihad (download: 332.13 MiB, generated: 1.49 GiB, post-processed: Unknown size, total: 1.82 GiB) to .../.cache/huggingface/datasets/arabic_billion_words/Alittihad/1.1.0/687a1f963284c8a766558661375ea8f7ab3fa3633f8cd9c9f42a53ebe83bfe17...
Downloading: 100%|███████████████████████████| 348M/348M [00:22<00:00, 15.8MB/s]
Dataset arabic_billion_words downloaded and prepared to .../.cache/huggingface/datasets/arabic_billion_words/Alittihad/1.1.0/687a1f963284c8a766558661375ea8f7ab3fa3633f8cd9c9f42a53ebe83bfe17. Subsequent calls will reuse this data.

## Actual results
Specify the actual results or traceback.

## Environment info

Machine 1:
- `datasets` version: 1.15.1
- Platform: Linux-5.8.0-63-generic-x86_64-with-glibc2.29
- Python version: 3.8.10
- PyArrow version: 4.0.1

Machine 2 (the bugged one)
- `datasets` version: 1.15.1
- Platform: Linux-4.4.0-210-generic-x86_64-with-glibc2.10
- Python version: 3.8.8
- PyArrow version: 6.0.0