Super slow iteration with trivial custom transform

### Describe the bug

Dataset is 10X slower when applying trivial transforms:

```
import time
import numpy as np
from datasets import Dataset, Features, Array2D

a = np.zeros((800, 800))
a = np.stack([a] * 1000)
features = Features({"a": Array2D(shape=(800, 800), dtype="uint8")})

ds1 = Dataset.from_dict({"a": a}, features=features).with_format('numpy')

def transform(batch):
return batch

ds2 = ds1.with_transform(transform)

%time sum(1 for _ in ds1)
%time sum(1 for _ in ds2)
```

```
CPU times: user 472 ms, sys: 319 ms, total: 791 ms
Wall time: 794 ms
CPU times: user 9.32 s, sys: 443 ms, total: 9.76 s
Wall time: 9.78 s
```

In my real code I'm using set_transform to apply some post-processing on-the-fly for the 2d array, but it significantly slows down the dataset even if the transform itself is trivial.

Related issue: https://github.com/huggingface/datasets/issues/5841

### Steps to reproduce the bug

Use code in the description to reproduce.

### Expected behavior

Trivial custom transform in the example should not slowdown the dataset iteration.

### Environment info

- `datasets` version: 2.18.0
- Platform: Linux-5.15.0-79-generic-x86_64-with-glibc2.35
- Python version: 3.11.4
- `huggingface_hub` version: 0.20.2
- PyArrow version: 15.0.0
- Pandas version: 1.5.3
- `fsspec` version: 2023.12.2