fix: support decimal-to-integer array casts

[XiangpengHao]

Summary

Decimal scalars can be cast to integers, but decimal arrays currently have no corresponding cast kernel. This makes expressions such as CAST(row_number() OVER () - 1 AS BIGINT) fail in downstream SQL execution with No CastKernel ... from decimal(21,0) to i64?.

Add decimal-array casts to all eight primitive integer types and share the conversion implementation with scalar casts. The policy remains: reject values outside the target range before truncating fractional digits toward zero. For example, 12.9 becomes 12, while 127.1 to i8 and -0.1 to an unsigned integer are rejected.

Changes

• Use checked integer arithmetic for scaling and range checks. This also corrects scalar rounding for large integers, handles negative scales, and supports i256 values without narrowing through i128 or f64.

• Reuse the existing decimal buffer conversion path, preserving validity and skipping invalid lanes.

• Add 64 regression cases covering scalar/array agreement, all six decimal storage widths, all integer targets, exact bounds, fractional overflow, nulls, empty arrays, and extreme scales.

• Add DataFusion and DuckDB SLTs over a decimal column stored in a Vortex file. DataFusion asserts cast pushdown and checks parity with pushdown disabled. DuckDB asserts that the cast stays in its own projection, preserving rounding semantics. Both cover nulls, fractions, large exact values, and BIGINT bounds.

Validation

• cargo nextest run -p vortex-array -E 'test(decimal)': 377 passed, including all 64 new cases.
• cargo nextest run -p vortex-sqllogictest -E 'test(decimal_to_integer)': 2 passed (DataFusion and DuckDB).
• Formatted vortex-array with the CI-pinned nightly-2026-09-10 formatter.
• Downstream driver rebuild and SQL verification: the failing expression now returns Int64 values 0, 1; the affected two-episode decode evaluation completed with 12 images.

Workspace-wide checks and clippy were not run locally.

[XiangpengHao]

@claude review

[myrrc]

LGTM. Can you also add a duckdb and a datafusion SLT please?

[codspeed-hq]

Merging this PR will degrade performance by 10.22%

⚠️ 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

⚡ 1 improved benchmark
❌ 5 regressed benchmarks
✅ 2191 untouched benchmarks
⏩ 218 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	15.2 µs	20.8 µs	-27.04%
❌	Simulation	random_i32[0.95]	91.8 µs	110.5 µs	-16.93%
❌	WallTime	mul_i64_nonnull_neon	17.2 µs	20.1 µs	-14.6%
❌	WallTime	multiply_shapes_neon[(16384, PerRowPerRow)]	17.2 µs	20 µs	-14.39%
❌	WallTime	mul_u32_nonnull_avx512	5.6 µs	6.3 µs	-10.83%
⚡	Simulation	random_i8[0.5]	94.2 µs	71 µs	+32.55%

Tip

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

---

_{Comparing XiangpengHao:codex/decimal-to-integer-cast (2f4101e) with develop (7ecae9e)}

Footnotes

1. 218 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. ↩