GPU: fuse dictionaries with nullable codes

[joseph-isaacs]

Independent of the other follow-up PRs; based directly on #9147. This draft contains one reviewable GPU correctness or performance concern.

What

• Pass dictionary length into the fused dispatch plan.
• Bounds-check unspecified physical codes at null positions in the CUDA gather.
• Keep validity propagation separate and enable fusion for nullable codes.

Why this is needed

This is not required by the current BtrBlocks GPU benchmark path. Vortex's normal DictEncoder intentionally stores nulls in the dictionary values so generated codes remain non-null. Nullable codes are nevertheless legal for externally supplied or transformed DictArray values; their physical null slots may contain arbitrary/out-of-range bits, so the bounds check is needed if CUDA claims support for that legal representation.

Recommendation: keep this as an independent general CUDA compatibility PR only if that scope is wanted; otherwise close/defer it rather than presenting it as a benchmark optimization.

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 degrade performance by 11.76%

⚠️ 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
✅ 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	-12.4%
❌	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-nullable-dictionary-dispatch (55cf62c) with develop (b825c4f)}

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. ↩