Loading HF Hub Dataset (private org repo) fails to load all features

### Describe the bug

When pushing a `Dataset` with multiple `Features` (`input`, `output`, `tags`) to Huggingface Hub (private org repo), and later downloading the `Dataset`, only `input` and `output` load - I believe the expected behavior is for all `Features` to be loaded by default?

### Steps to reproduce the bug

Pushing the data. `data_concat` is a `list` of `dict`s.
```python
for datum in data_concat:
datum_tags = {d["key"]: d["value"] for d in datum["tags"]}
split_fraction = # some logic that generates a train/test split number
if split_faction < test_fraction:
data_test.append(datum)
else:
data_train.append(datum)

dataset = DatasetDict(
{
"train": Dataset.from_list(data_train),
"test": Dataset.from_list(data_test),
"full": Dataset.from_list(data_concat),
},
)

dataset_shuffled = dataset.shuffle(seed=shuffle_seed)

dataset_shuffled.push_to_hub(
repo_id=hf_repo_id,
private=True,
config_name=m,
revision=revision,
token=hf_token,
)
```

Loading it later:
```python
dataset = datasets.load_dataset(
path=hf_repo_id,
name=name,
token=hf_token,
)
```

Produces:
```
DatasetDict({
train: Dataset({
features: ['input', 'output'],
num_rows:
})
test: Dataset({
features: ['input', 'output'],
num_rows:
})
full: Dataset({
features: ['input', 'output'],
num_rows:
})
})
```

### Expected behavior

The expected result is below:
```
DatasetDict({
train: Dataset({
features: ['input', 'output', 'tags'],
num_rows:
})
test: Dataset({
features: ['input', 'output', 'tags'],
num_rows:
})
full: Dataset({
features: ['input', 'output', 'tags'],
num_rows:
})
})
```

My workaround is as follows:
```python
dsinfo = datasets.get_dataset_config_info(
path=data_files,
config_name=data_config,
token=hf_token,
)

allfeatures = dsinfo.features.copy()
if "tags" not in allfeatures:
allfeatures["tags"] = [{"key": Value(dtype="string", id=None), "value": Value(dtype="string", id=None)}]

dataset = datasets.load_dataset(
path=data_files,
name=data_config,
features=allfeatures,
token=hf_token,
)
```

Interestingly enough (and perhaps a related bug?), if I don't add the `tags` to `allfeatures` above (i.e. only loading `input` and `output`), it throws an error when executing `load_dataset`:

```
ValueError: Couldn't cast
tags: list>
child 0, element: struct
child 0, key: string
child 1, value: string
input:
output:
-- schema metadata --
huggingface: '{"info": {"features": {"tags": [{"key": {"dtype": "string",' + 532
to
{'input': , 'output':
because column names don't match
```

Traceback for this:
```
Traceback (most recent call last):
File "", line 1, in
File "/Users/bt/github/core/.venv/lib/python3.11/site-packages/datasets/load.py", line 2152, in load_dataset
builder_instance.download_and_prepare(
File "/Users/bt/github/core/.venv/lib/python3.11/site-packages/datasets/builder.py", line 948, in download_and_prepare
self._download_and_prepare(
File "/Users/bt/github/core/.venv/lib/python3.11/site-packages/datasets/builder.py", line 1043, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/Users/bt/github/core/.venv/lib/python3.11/site-packages/datasets/builder.py", line 1805, in _prepare_split
for job_id, done, content in self._prepare_split_single(
File "/Users/bt/github/core/.venv/lib/python3.11/site-packages/datasets/builder.py", line 1950, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.builder.DatasetGenerationError: An error occurred while generating the dataset
```

### Environment info

- `datasets` version: 2.15.0
- Platform: macOS-14.0-arm64-arm-64bit
- Python version: 3.11.5
- `huggingface_hub` version: 0.19.4
- PyArrow version: 14.0.1
- Pandas version: 2.1.4
- `fsspec` version: 2023.10.0