GPU: remove FSST metadata upload callbacks

[joseph-isaacs]

Stacked only on #9426. This draft contains one reviewable GPU correctness or performance concern.

What

• Submit FSST metadata copies directly on the execution stream.
• Remove host callbacks used only to stage those small uploads.

Why this is needed

The performance goal is real: the generic async upload future installs a host callback and waits for all preceding stream work just to keep tiny upload sources alive. The FSST symbol and length tables are ordinary pageable vectors, so they do not need that callback-heavy lifetime path.

However, this draft is not merge-ready as written. It also changes the validity-buffer upload to ensure_on_device_sync. Vortex's “sync” helper currently calls cudarc's stream memcpy_htod, which uses cuMemcpyHtoDAsync; cudarc documents that the source must not move until completion. A validity source may be pinned, so dropping it immediately can be unsafe. Keep validity on the awaited lifetime-safe path, or use a truly synchronous driver copy, and restrict the callback removal to proven-pageable metadata.

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 improve performance by 10.18%

⚠️ 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
✅ 2042 untouched benchmarks
⏩ 46 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	take[core/shuffled/primitive/nonnull/chunks=1024/indices=10000]	14.2 ms	12.9 ms	+10.18%

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing joe/gpu-fsst-no-metadata-callbacks (6cb8843) with joe/gpu-fsst-device-output (b215956)}

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. ↩