[CI][ROCm] Make hf-xet reconstruction safe on shared NFS

[AndreasKaratzas]

Summary

• Detect an NFS-backed HF_HOME in the AMD hardware test runner.
• Keep Xet and the persistent Hub cache enabled while moving ephemeral Xet state to local storage and disabling high-performance vectored reconstruction only on NFS.

hf-xet 1.5.1 stalled on the large vectored write after the first tokenizer fragment, and later jobs convoyed behind the Hub lock. Non-NFS jobs are unchanged.

Validation

• bash -n .buildkite/scripts/hardware_ci/run-amd-test.sh
• Changed-file pre-commit passed.
• Local syscall controls reconstructed Apeach and Qwen3-Next with scalar writes. Cold concurrent validation on Buildkite NFS remains pending.

Buildkite: Entrypoints Integration (Pooling)

Duplicate check: no open PR was found for hf-xet NFS.

AI assistance: Codex assisted with investigation, implementation, and validation; Andreas directed the design and reviewed the resulting changes.