feat: add generic Hugging Face dataset converter

[joseph-isaacs]

Summary

• add a resumable converter for Parquet shards in any Hugging Face dataset repository
• support Parquet Zstd, Vortex, and compact Vortex outputs with deterministic destination paths
• require Xet-backed source and destination repositories and expose bounded Xet download concurrency
• bound prefetched source data and pending upload data with independent worker pools
• batch Hub commits, verify downloads, checkpoint progress, and delete local outputs only after successful commits
• add a Linux process-tree resource monitor for CPU, memory, disk, and network usage
• redirect monitored child output to a durable log and make progress logging safe after terminal detachment
• distinguish transfer failures from logging failures and reuse local encodings during checkpoint resume
• preserve checkpoint compatibility when operational Xet cache and concurrency settings change

Validation

• data/fineweb-convert/.venv/bin/python -m unittest scripts/tests/test_hf_parquet_convert.py (17 tests)
• python3.12 -m py_compile scripts/huggingface-convert.py scripts/monitor-process.py scripts/tests/test_hf_parquet_convert.py
• closed-stdout end-to-end monitor test with durable child-log verification
• live checkpoint resume: skipped 120 committed shards and reused pending local encodings
• converter CLI help inspection
• resource-monitor smoke test
• git diff --check

The converter has also been exercised against a large FineWeb upload, but it contains no FineWeb-specific repository or path assumptions.

[codspeed-hq]

Merging this PR will regress 1 benchmark

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 3 improved benchmarks
❌ 1 regressed benchmark
✅ 2126 untouched benchmarks
⏩ 106 skipped benchmarks¹
🗄️ 4 archived benchmarks run²

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	mul_u32_nonnull_avx512	5.6 µs	6.3 µs	-11.87%
⚡	WallTime	arrow_checked_add_u32_avx2[16384]	21.3 µs	17.7 µs	+20.53%
⚡	Simulation	take[duplicates/repeated/primitive/nonnull/chunks=16/indices=1000]	245.7 µs	205.4 µs	+19.61%
⚡	WallTime	arrow_checked_add_u32_neon[16384]	13.5 µs	12.2 µs	+10.52%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing ji/hf-dataset-vortex-converter (59f79b4) with develop (59a056d)}

Footnotes

1. 106 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

2. 4 benchmarks were run, but are now archived. If they were deleted in another branch, consider rebasing to remove them from the report. Instead if they were added back, click here to restore them. ↩