GPU: pack FSST metadata into one upload

[joseph-isaacs]

Stacked only on #9430 (which itself depends on #9426). This draft contains one reviewable GPU correctness or performance concern.

What

• Pack the 255 FSST symbols and lengths into one C-compatible table.
• Allocate and upload one device buffer instead of two per FSST decode.
• Keep the measured cached-global-memory lookup strategy.

Why this is needed

This is a small incremental FSST cleanup, not the full metadata solution. Symbols and lengths are always consumed together, so packing them saves one allocation and one H2D API call per FSST decode. It does not remove the remaining per-decode metadata upload or implement the requested scan-scoped table cache.

Recommendation: fold this into the corrected #9430 or defer it behind scan-scoped FSST metadata reuse; the standalone benefit is modest.

Validation

Checked independently against this PR's current base:

• cargo check -p vortex-cuda --all-features

The combined implementation also passed Rust/CUDA formatting, 164 focused dynamic-dispatch tests, and 28 focused constant-array tests.

[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.

⚡ 1 improved benchmark
❌ 1 regressed benchmark
✅ 2041 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_scalar[65536]	8.2 µs	9.4 µs	-12.16%
⚡	Simulation	decompress[u64, (10000, 4)]	402.6 µs	310.6 µs	+29.62%

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-pack-fsst-metadata (b7bf156) with joe/gpu-fsst-no-metadata-callbacks (6cb8843)}

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. ↩