[DO NOT MERGE] Add SIMD primitive comparison kernels

[connortsui20]

Summary

Moves the explicit AVX2 and AVX-512 primitive comparison kernels from #9547 into the handwritten comparison path. This leaves the RowFn work in #9547 and #9548 untouched while preserving the existing portable lane-kernel fallback on non-x86 targets and x86-64 CPUs without AVX2.

Top of a three-PR stack: #9598 adds the comparison benchmarks and their #[cpu_features] tags, #9599 sweeps the same tag across the rest of the numeric arithmetic and comparison microbenchmarks, and this PR adds the kernels. The benchmarks land first so the walltime legs record the portable baseline on each feature set, and this PR's CodSpeed report compares kernels against it rather than introducing both at once.

Changes

• Adds full-word SIMD kernels for all primitive types, comparison operators, and array/constant orientations, with scalar tail handling, split by lane width.
• Adds boundary and nullable coverage.
• Confirms the intended x86-64 codegen by cross-compiling for macOS: AVX2 uses packed ymm masks and AVX-512 uses zmm/k masks with direct bitmap stores. This environment cannot provide x86 runtime timings.

[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.

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 6 improved benchmarks
❌ 1 regressed benchmark
✅ 1934 untouched benchmarks
🆕 156 new benchmarks
⏩ 106 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	compare_int_constant_left_avx2	3 µs	3.4 µs	-10.88%
⚡	WallTime	compare_u8_avx2	3.4 µs	1.7 µs	×2
⚡	WallTime	compare_f32_avx2	4.6 µs	3.1 µs	+48.27%
⚡	WallTime	compare_u8_avx512	2.3 µs	1.6 µs	+46.71%
⚡	WallTime	compare_f32_avx512	3.6 µs	2.7 µs	+37.02%
⚡	Simulation	search_index_above_max_chunked	608.7 µs	549.4 µs	+10.79%
⚡	Simulation	search_index_in_range_chunked	610.3 µs	550.9 µs	+10.77%
🆕	WallTime	add_i32_nonnull_neon	N/A	7.7 µs	N/A
🆕	WallTime	add_i64_constant_neon	N/A	9.8 µs	N/A
🆕	WallTime	add_i64_nonnull_neon	N/A	9.8 µs	N/A
🆕	WallTime	add_i64_nullable_neon	N/A	11.4 µs	N/A
🆕	WallTime	add_shapes_neon[(128, ConstantPerRow)]	N/A	3.1 µs	N/A
🆕	WallTime	add_shapes_neon[(128, PerRowConstant)]	N/A	3.1 µs	N/A
🆕	WallTime	add_shapes_neon[(128, PerRowNullableConstant)]	N/A	3.7 µs	N/A
🆕	WallTime	add_shapes_neon[(128, PerRowPerRow)]	N/A	1.9 µs	N/A
🆕	WallTime	add_shapes_neon[(16384, ConstantPerRow)]	N/A	9.8 µs	N/A
🆕	WallTime	add_shapes_neon[(16384, PerRowConstant)]	N/A	9.9 µs	N/A
🆕	WallTime	add_shapes_neon[(16384, PerRowNullableConstant)]	N/A	10.5 µs	N/A
🆕	WallTime	add_shapes_neon[(16384, PerRowPerRow)]	N/A	9.8 µs	N/A
🆕	WallTime	add_u32_nonnull_neon	N/A	6.6 µs	N/A
...	...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

Tip

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

---

_{Comparing ct/primitive-comparison-simd (1163162) with develop (0ef496c)²}

Footnotes

1. 106 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. ↩

2. No successful run was found on develop (17bd9e2) during the generation of this report, so 0ef496c was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩