GPU benchmark: add reusable decompression profiling metrics

[joseph-isaacs]

Based directly on develop; it has no dependency on the Parquet benchmark PR.

What

• Add opt-in VORTEX_GPU_PROFILE=wall|gpu diagnostics to the existing GPU Vortex benchmark.
• Emit one JSON record with file/layout sizes, row/batch/dispatch counts, open/scan/read/dispatch/final-sync wall times, and timings grouped by encoding tree.
• In gpu mode, bracket field dispatches with CUDA events and report device-stream time.
• Preserve the original decode loop as an early-return fast path when profiling is disabled.

Why this is needed

This is measurement infrastructure, not a speedup. It lets us separate host planning/dispatch, I/O/backpressure, final synchronization, and actual device execution without editing and rebuilding the benchmark for each investigation. Keeping the unprofiled path intact also prevents the measurement tooling from changing normal benchmark results.

Validation

• cargo +nightly fmt --all
• cargo check -p compress-bench --features cuda,unstable_encodings
• git diff --check

[codspeed-hq]

Merging this PR will degrade performance by 11.88%

⚠️ 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.

❌ 2 regressed benchmarks
✅ 1998 untouched benchmarks
⏩ 89 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	words_gather_scalar[65536]	8.2 µs	9.4 µs	-12.65%
❌	WallTime	words_gather_dispatch[1024]	8 ns	9 ns	-11.11%

Tip

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

---

_{Comparing joe/gpu-decompression-profiling (b6cf956) with develop (b825c4f)}

Footnotes

1. 89 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. ↩