bug(services/hf): hf_dataset behavior tests timeout on main

[Xuanwo]

Describe the bug

core / ubuntu-latest / hf / hf_dataset failed on main in the Behavior Test workflow:

• run: https://github.com/apache/opendal/actions/runs/26216207614
• job: https://github.com/apache/opendal/actions/runs/26216207614/job/77139509964
• head sha: 10975e960a18e9133b66683ebc2a7f4ad69d2885
• trigger commit: docs: update architecture overview image (#7574)

The triggering commit only changed README.md and website/static/img/architectural.png, so this does not look caused by an HF service code change in that commit.

The job created a temporary private dataset repo successfully:

• OPENDAL_HF_REPO_ID=opendal/test-dataset-26216207614-test-b4d1f2ca
• OPENDAL_HF_REPO_TYPE=dataset
• OPENDAL_TEST=hf

The behavior run then hit repeated 10s I/O timeouts against the real HF dataset backend. The failures were spread across write/delete/list-related tests instead of one assertion:

will retry Write (attempt 1) after 1s because: Unexpected (temporary) at write => io operation timeout reached
Context:
   timeout: 10

will retry Delete (attempt 1) after 1s because: Unexpected (temporary) at delete => io operation timeout reached
Context:
   timeout: 10

failures:
    behavior::test_read_full
    behavior::test_batch_delete
    behavior::test_list_file_with_recursive
    behavior::test_list_dir_with_file_path

test result: FAILED. 88 passed; 4 failed; 0 ignored; 0 measured; 0 filtered out; finished in 119.18s

Post-job cleanup also failed:

Cleanup failed: HTTP 403: {"error":"You have read access but not the required permissions for this operation"}

This suggests we should inspect both the HF dataset behavior-test setup and the token/repo-permission model. It may also be backend slowness/rate-limiting or our 10s timeout being too aggressive for HF/XET writes under this test shape.

cc @kszucs since you authored the recent HF/XET write path and related HF fixes.

Steps to Reproduce

Run the core behavior test matrix for the HF dataset setup on Linux:

# In CI this is generated by .github/workflows/test_behavior_core.yml
# via .github/services/hf/hf_dataset/action.yml.
OPENDAL_TEST=hf \
OPENDAL_HF_REPO_TYPE=dataset \
OPENDAL_HF_REPO_ID=<temporary-private-dataset-repo> \
OPENDAL_HF_TOKEN=<token> \
RUST_TEST_THREADS=1 \
cargo test -p opendal --features services-hf,tests behavior

The observed failure happened in GitHub Actions on Ubuntu 24.04 with Rust 1.95.0.

Expected Behavior

The HF dataset behavior test should pass reliably, or fail with a clear HF/OpenDAL error that identifies the real backend/token problem. It should not fail multiple unrelated behavior tests via generic 10s I/O timeouts.

Additional Context

Relevant local files:

• .github/services/hf/hf_dataset/action.yml
• .github/actions/hf-temp-repo/setup.js
• .github/actions/hf-temp-repo/cleanup.js
• core/services/hf/

Related but different issue: #7367 tracked the Java blocking HF/XET segfault and was mitigated by disabling Java HF behavior tests. This issue is about the core Rust HF dataset behavior job timing out on main.