ArrayXD with None as leading dim incompatible with DatasetCardData

### Describe the bug

Creating a dataset with ArrayXD features leads to errors when downloading from hub due to DatasetCardData removing the Nones

@lhoestq

### Steps to reproduce the bug

```python
import numpy as np
from datasets import Array2D, Dataset, Features, load_dataset

def examples_generator():
for i in range(4):
yield {
"array_1d": np.zeros((10,1), dtype="uint16"),
"array_2d": np.zeros((10, 1), dtype="uint16"),
}

features = Features(array_1d=Array2D((None,1), "uint16"), array_2d=Array2D((None, 1), "uint16"))
dataset = Dataset.from_generator(examples_generator, features=features)
dataset.push_to_hub("alex-hh/test_array_1d2d")
ds = load_dataset("alex-hh/test_array_1d2d")
```

Source of error appears to be DatasetCardData.to_dict invoking DatasetCardData._remove_none
```python
from huggingface_hub import DatasetCardData
from datasets.info import DatasetInfosDict

dataset_card_data = DatasetCardData()
DatasetInfosDict({"default": dataset.info.copy()}).to_dataset_card_data(dataset_card_data)
print(dataset_card_data.to_dict()) # removes Nones in shape
```

### Expected behavior

Should be possible to load datasets saved with shape None in leading dimension

### Environment info

3.0.2 and latest huggingface_hub