Out of memory even though using load_dataset(..., streaming=True)

### Describe the bug

I am iterating over a large dataset that I load using streaming=True to avoid running out of memory. Unfortunately, I am observing that memory usage increases over time and I'm finally running in an oom.

### Steps to reproduce the bug

```
ds = load_dataset("openslr/librispeech_asr", split="train.clean.360", streaming=True)
for i,sample in enumerate(tqdm(ds)):
target_file = os.path.join(NSFW_TARGET_FOLDER, f'audio{i}.wav')
try:
sf.write(target_file, sample['audio']['array'], samplerate=sample['audio']['sampling_rate'])
except Exception as e:
print(f"Could not write audio {i} in ds: {e}")
```

### Expected behavior

I'd expect to have a small memory footprint and memory being freed after each iteration of the for loop. Instead the memory usage is increasing. I tried to remove the logic to write the sound file and just print the sample but the issue remains the same.

### Environment info

Python 3.12.11
Ubuntu 24
datasets 4.0.0 and 3.6.0