Offset overflow when executing Dataset.map

### Describe the bug

Hi, I'm trying to use `.map` method to cache multiple random crops from the image to speed up data processing during training, as the image size is too big.

The map function executes all iterations, and then returns the following error:

```bash
Traceback (most recent call last):
File "/home/ubuntu/miniconda3/envs/enhancement/lib/python3.8/site-packages/datasets/arrow_dataset.py", line 3353, in _map_single
writer.finalize() # close_stream=bool(buf_writer is None)) # We only close if we are writing in a file
File "/home/ubuntu/miniconda3/envs/enhancement/lib/python3.8/site-packages/datasets/arrow_writer.py", line 582, in finalize
self.write_examples_on_file()
File "/home/ubuntu/miniconda3/envs/enhancement/lib/python3.8/site-packages/datasets/arrow_writer.py", line 446, in write_examples_on_file
self.write_batch(batch_examples=batch_examples)
File "/home/ubuntu/miniconda3/envs/enhancement/lib/python3.8/site-packages/datasets/arrow_writer.py", line 555, in write_batch
self.write_table(pa_table, writer_batch_size)
File "/home/ubuntu/miniconda3/envs/enhancement/lib/python3.8/site-packages/datasets/arrow_writer.py", line 567, in write_table
pa_table = pa_table.combine_chunks()
File "pyarrow/table.pxi", line 3315, in pyarrow.lib.Table.combine_chunks
File "pyarrow/error.pxi", line 144, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 100, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: offset overflow while concatenating arrays
```

Here is the minimal code (`/home/datasets/DIV2K_train_HR` is just a folder of images that can be replaced by any appropriate):

### Steps to reproduce the bug

```python
from glob import glob

import torch
from datasets import Dataset, Image
from torchvision.transforms import PILToTensor, RandomCrop

file_paths = glob("/home/datasets/DIV2K_train_HR/*")

to_tensor = PILToTensor()
crop_transf = RandomCrop(size=256)

def prepare_data(example):
tensor = to_tensor(example["image"].convert("RGB"))
return {"hr": torch.stack([crop_transf(tensor) for _ in range(25)])}

train_data = Dataset.from_dict({"image": file_paths}).cast_column("image", Image())
train_data = train_data.map(
prepare_data,
cache_file_name="/home/datasets/DIV2K_train_HR_crops.tmp",
desc="Caching multiple random crops of image",
remove_columns="image",
)
print(train_data[0].keys(), train_data[0]["hr"].shape)
```

### Expected behavior

Cached file is stored at `"/home/datasets/DIV2K_train_HR_crops.tmp"`, output is `dict_keys(['hr']) torch.Size([25, 3, 256, 256])`

### Environment info

- `datasets` version: 2.10.1
- Platform: Linux-5.15.0-67-generic-x86_64-with-glibc2.10
- Python version: 3.8.16
- PyArrow version: 11.0.0
- Pandas version: 1.5.3
- Pytorch version: 2.0.0+cu117
- torchvision version: 0.15.1+cu117