`_resolve_features` ignores the token

Describe the bug

When calling _resolve_features() on a gated dataset, ie. a dataset which requires a token to be loaded, the token seems to be ignored even if it has been provided to load_dataset before.

Steps to reproduce the bug

import os

os.environ["HF_ENDPOINT"] = "https://hub-ci.huggingface.co/"
hf_token = "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"

from datasets import load_dataset

# public
dataset_name = "__DUMMY_DATASETS_SERVER_USER__/repo_csv_data-16612654226756"
config_name = "__DUMMY_DATASETS_SERVER_USER__--repo_csv_data-16612654226756"
split_name = "train"

iterable_dataset = load_dataset(
    dataset_name,
    name=config_name,
    split=split_name,
    streaming=True,
    use_auth_token=hf_token,
)
iterable_dataset = iterable_dataset._resolve_features()
print(iterable_dataset.features)

# gated
dataset_name = "__DUMMY_DATASETS_SERVER_USER__/repo_csv_data-16612654317644"
config_name = "__DUMMY_DATASETS_SERVER_USER__--repo_csv_data-16612654317644"
split_name = "train"


iterable_dataset = load_dataset(
    dataset_name,
    name=config_name,
    split=split_name,
    streaming=True,
    use_auth_token=hf_token,
)
try:
    iterable_dataset = iterable_dataset._resolve_features()
except FileNotFoundError as e:
    print("FAILS")

Expected results

I expect to have the same result on a public dataset and on a gated (or private) dataset, if the token has been provided.

Actual results

An exception is thrown on gated datasets.

Environment info

• datasets version: 2.4.0
• Platform: Linux-5.15.0-1017-aws-x86_64-with-glibc2.35
• Python version: 3.9.6
• PyArrow version: 7.0.0
• Pandas version: 1.4.2

[lhoestq]

yes definitely !

[lhoestq]

Yes exactly, this is a known bug