xopen: invalid start byte for streaming dataset with trust_remote_code=True

### Describe the bug

I am trying to load YODAS2 dataset with datasets==3.6.0
```
from datasets import load_dataset
next(iter(load_dataset('espnet/yodas2', name='ru000', split='train', streaming=True, trust_remote_code=True)))
```
And get `UnicodeDecodeError: 'utf-8' codec can't decode byte 0xa8 in position 1: invalid start byte`

The cause of the error is the following:

```
from datasets.utils.file_utils import xopen

filepath = 'https://huggingface.co/datasets/espnet/yodas2/resolve/c9674490249665d658f527e2684848377108d82c/data/ru000/text/00000000.json'
xopen(filepath, 'r').read()
>>> UnicodeDecodeError: 'utf-8' codec can't decode byte 0xa8 in position 1: invalid start byte
```

And the cause of this is the following:

```
import fsspec

fsspec.open(
'hf://datasets/espnet/yodas2@c9674490249665d658f527e2684848377108d82c/data/ru000/text/00000000.json',
mode='r',
hf={'token': None, 'endpoint': 'https://huggingface.co'},
).open().read()
>>> UnicodeDecodeError: 'utf-8' codec can't decode byte 0xa8 in position 1: invalid start byte
```

Is it true that streaming=True loading is not supported anymore for trust_remote_code=True, even with datasets==3.6.0? This breaks backward compatibility.

### Steps to reproduce the bug

```
from datasets import load_dataset
next(iter(load_dataset('espnet/yodas2', name='ru000', split='train', streaming=True)))
```

### Expected behavior

No errors expected

### Environment info

datasets==3.6.0, ubuntu 24.04