download_mode="force_redownload" does not refresh cached dataset

### Describe the bug

`load_datasets` does not refresh dataset when features are imported from external file, even with `download_mode="force_redownload"`. The bug is not limited to nested fields, however it is more likely to occur with nested fields.

### Steps to reproduce the bug

To reproduce the bug 3 files are needed: `dataset.py` (contains dataset loading script), `schema.py` (contains features of dataset) and `main.py` (to run `load_datasets`)

`dataset.py`
```python
import datasets
from schema import features

class NewDataset(datasets.GeneratorBasedBuilder):

def _info(self):
return datasets.DatasetInfo(
features=features
)

def _split_generators(self, dl_manager):
return [
datasets.SplitGenerator(
name=datasets.Split.TRAIN
)
]

def _generate_examples(self):
data = [
{"id": 0, "nested": []},
{"id": 1, "nested": []}
]
for key, example in enumerate(data):
yield key, example
```

`schema.py`
```python
import datasets
features = datasets.Features(
{
"id": datasets.Value("int32"),
"nested": [
{"text": datasets.Value("string")}
]
}
)
```

`main.py`
```python
import datasets
a = datasets.load_dataset("dataset.py")
print(a["train"].info.features)
```

Now if `main.py` is run it prints the following correct output: `{'id': Value(dtype='int32', id=None), 'nested': [{'text': Value(dtype='string', id=None)}]}`. However, if f.e. the label of the feature "text" is changed to something else, f.e. to

`schema.py`
```python
import datasets
features = datasets.Features(
{
"id": datasets.Value("int32"),
"nested": [
{"textfoo": datasets.Value("string")}
]
}
)
```

`main.py` still prints `{'id': Value(dtype='int32', id=None), 'nested': [{'text': Value(dtype='string', id=None)}]}`, even if run with `download_mode="force_redownload"`. The only fix is to delete the folder in the cache.

### Expected behavior

The cached dataset is deleted and refreshed when using `load_datasets` with `download_mode="force_redownload"`.

### Environment info

- `datasets` version: 2.7.0
- Platform: Windows-10-10.0.19041-SP0
- Python version: 3.7.9
- PyArrow version: 10.0.0
- Pandas version: 1.3.5