flatten_indices doesn't work with pandas format

### Describe the bug

Hi,
I noticed that `flatten_indices` throws an error when the batch format is `pandas`. This is probably due to the fact that flatten_indices uses map internally which doesn't accept dataframes as the transformation function output

### Steps to reproduce the bug

tabular_data = pd.DataFrame(np.random.randn(10,10))
tabular_data = datasets.arrow_dataset.Dataset.from_pandas(tabular_data)
tabular_data.with_format("pandas").select([0,1,2,3]).flatten_indices()

### Expected behavior

No error thrown

### Environment info

- `datasets` version: 2.10.1
- Python version: 3.9.5
- PyArrow version: 11.0.0
- Pandas version: 1.4.1