bench: run the full compress suite under --gpu-decompress

[joseph-isaacs]

Rationale for this change

--gpu-decompress in compress-bench was gated behind an incremental allow-list that
contained a single entry (TPC-H l_comment canonical), so the GPU compression benchmark
job only ever measured one dataset. This widens GPU coverage to the whole compress suite,
including the airquality dataset.

Opened as a draft to let the GPU benchmark job tell us which datasets actually have
working CUDA decode kernels — that verification is exactly what the allow-list was
standing in for, and it cannot be done locally without a GPU.

What changes are included in this PR?

• Remove the gpu_decompress_benchmarks allow-list from benchmarks/compress-bench/src/main.rs.
  With no --datasets filter, --gpu-decompress now runs every dataset in the suite:
  taxi, the Public BI sets (Arade, Bimbo, CMSprovider, Euro2016, Food, HashTags), both
  TPC-H l_comment variants, the StructListOfInts variants, plus airquality and rplace.
• The CPU default path is unchanged and still skips airquality/rplace. Note that the GPU
  path now downloads both from pcodec's public S3 bucket on every run; the existing CPU-side
  exclusion exists to avoid creating egress charges for pcodec, so this may want caching
  rather than a permanent CI cost.
• Update benchmarks/compress-bench/README.md to match.

What APIs are changed? Are there any user-facing changes?

No library API changes. The only behavior change is to the compress-bench benchmark binary:
--gpu-decompress no longer filters the dataset list.

Checks

• cargo clippy -p compress-bench --all-targets passes.
• cargo +nightly fmt could not be run in this environment (the nightly toolchain ICEs);
  the edits are formatting-neutral but CI fmt is unverified.
• The GPU run itself is unverified locally — no CUDA device available. That is what this
  draft's benchmark job is for.

---

Generated by Claude Code

[codspeed-hq]

Merging this PR will degrade performance by 12.14%

⚠️ 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.

⚡ 1 improved benchmark
❌ 3 regressed benchmarks
✅ 2033 untouched benchmarks
⏩ 46 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	words_gather_dispatch[1024]	9 ns	13 ns	-30.77%
❌	WallTime	words_gather_scalar[65536]	8.3 µs	9.4 µs	-11.83%
❌	Simulation	take[small_m/shuffled/primitive/nonnull/chunks=16384/indices=16]	1 ms	1.2 ms	-11.26%
⚡	WallTime	words_gather_dispatch[1024]	33 ns	30 ns	+10%

Tip

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

---

_{Comparing claude/gpu-compress-benchmarks-x788xc (cf631cd) with develop (1a493b6)}

Footnotes

1. 46 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. ↩

[github-actions]

BENCHMARK FAILED

GPU Compression failed. Check the workflow run for details.