Bug: Can't download TriviaQA with `load_dataset` - custom `cache_dir`

Hello,
I'm having issue downloading TriviaQA dataset with `load_dataset`.

## Environment info
- `datasets` version: 1.1.3
- Platform: Linux-4.19.129-aufs-1-x86_64-with-debian-10.1
- Python version: 3.7.3

## The code I'm running:
```python
import datasets
dataset = datasets.load_dataset("trivia_qa", "rc", cache_dir = "./datasets")
```

## The output:
1. Download begins:
```
Downloading and preparing dataset trivia_qa/rc (download: 2.48 GiB, generated: 14.92 GiB, post-processed: Unknown size, total: 17.40 GiB) to /cs/labs/gabis/sapirweissbuch/tr
ivia_qa/rc/1.1.0/e734e28133f4d9a353af322aa52b9f266f6f27cbf2f072690a1694e577546b0d...
Downloading: 17%|███████████████████▉ | 446M/2.67G [00:37<04:45, 7.77MB/s]
```
2. 100% is reached
3. It got stuck here for about an hour, and added additional 30G of data to "./datasets" directory. I killed the process eventually.

A similar issue can be observed in Google Colab:

https://colab.research.google.com/drive/1nn1Lw02GhfGFylzbS2j6yksGjPo7kkN-?usp=sharing

## Expected behaviour:
The dataset "TriviaQA" should be successfully downloaded.