Add vortex.is_nan expression with stats-based pruning

[Ecthlion]

Summary

• Tracking Issue: Expressions missing that could improve pushdown #9913

Adds the IS NAN expression called out in #9913, implemented as a strict RowFn as suggested in the issue: null inputs propagate to null outputs (output nullability follows the input), and only primitive float inputs are accepted.

Changes

• New vortex.is_nan scalar function in vortex-array implemented with the RowFn row-wise kernel framework, registered in the scalar fn session and exposed as expr::is_nan / expr::bound::is_nan.
• Stats rewrites mirroring the is_null rules so the predicate prunes from zone maps:
  • falsify: nan_count(x) = 0 or all_non_nan(x)
  • satisfy: nan_count(x) = row_count() or all_nan(x)
• Unit tests for evaluation (valid/null floats, constants, sliced arrays, non-float rejection), dtype strictness, display, and the falsify/satisfy rewrites.

AI disclosure: this change was developed with assistance from Kimi K3.

[codspeed]

Merging this PR will degrade performance by 10.44%

⚠️ 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

⚡ 2 improved benchmarks
❌ 10 regressed benchmarks
✅ 2196 untouched benchmarks
⏩ 293 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	mul_u64_nonnull_neon	28.9 µs	40 µs	-27.91%
❌	Simulation	take_fsl_u32_random[16, 100]	125.9 µs	164.7 µs	-23.51%
❌	Simulation	bench_compare_sliced_dict_primitive[(3333, 10000)]	160.6 µs	197.9 µs	-18.87%
❌	WallTime	dbp_assemble_kernel_avx2[(I128, 1024)]	464 ns	560 ns	-17.14%
❌	WallTime	mul_i64_nonnull_neon	32.8 µs	39 µs	-16.11%
❌	WallTime	multiply_shapes_neon[(32768, PerRowPerRow)]	32.7 µs	38.9 µs	-15.94%
❌	WallTime	dbp_assemble_kernel_avx512[(I128, 1024)]	463 ns	544 ns	-14.89%
❌	WallTime	dict_canonicalize_gt_u8_neon[1000000]	489.7 µs	570.9 µs	-14.22%
❌	WallTime	filtered_owned_i64_avx2[OneNullInEight]	22.6 µs	25.7 µs	-12.18%
❌	WallTime	dbp_assemble_kernel_narrow_msp_avx2[(I128, 1024)]	629 ns	709 ns	-11.28%
⚡	Simulation	take_fsl_f16_random[16, 100]	179.5 µs	118.2 µs	+51.81%
⚡	WallTime	filtered_sink_i64_avx512[OneNullInEight]	26.5 µs	22.5 µs	+17.91%

Tip

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

---

_{Comparing Ecthlion:add-is-nan-expression (265eb6d) with develop (a542cbd)}

Footnotes

1. 293 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. ↩