Add native decimal Add/Sub kernels

[mhk197]

Adds shared infrastructure for Decimal arithmetic and Add/Sub kernels.

Semantics

Both operands must share one DecimalDType (as compare requires since #8661; coerce_args/least_supertype already coerce to a common decimal dtype), and the result keeps that dtype with unioned nullability. Add/Sub apply to the unscaled stored integers — exact at a shared scale.

Implementation

• numeric.rs splits into numeric/{mod,primitive,decimal,tests}.rs, mirroring the compare split in Implement compare in vortex instead of falling back to arrow #8661. The generic checked lane loops move to numeric/mod.rs with bounds relaxed to T: Default so wider decimal types qualify.
• numeric/decimal.rs follows compare/decimal.rs: operands canonicalize through execute::<DecimalArray> (so Chunked, DecimalByteParts, Dict, … work with no kernel registration) or extract constants, widen once to a working type, and run one-pass checked lanes.
• The working width is max(storage widths, smallest type for p+1 digits), so in-precision inputs cannot spuriously overflow the sum, and the range check subsumes width overflow. Note precision can be stricter than width: 60 + 60 fits an i8 but exceeds precision 2 and errors.
• The result keeps the working width; DecimalData explicitly supports wider-than-necessary storage. Happy to add a BigCast narrowing pass if preferred.
• The binary-numeric conformance harness accepts decimals, checking Add/Sub element-wise against DecimalScalar::checked_binary_numeric (whose Add/Sub are already correct) for constants 0, 1, -1, 10^s, MAX_BY_PRECISION[p] in both operand orders; wired into the decimal, chunked, and decimal-byte-parts compute tests.
• widened_buffer moves from compare/decimal.rs to arrays/decimal so compare and numeric share it.

[codspeed-hq]

Merging this PR will improve performance by 12.73%

⚡ 6 improved benchmarks
❌ 1 regressed benchmark
✅ 1630 untouched benchmarks
🆕 2 new benchmarks
⏩ 38 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	rebuild_naive	91.3 µs	109.3 µs	-16.53%
⚡	Simulation	bitwise_not_vortex_buffer_mut[128]	244.4 ns	186.1 ns	+31.34%
⚡	Simulation	bitwise_not_vortex_buffer_mut[1024]	304.7 ns	246.4 ns	+23.68%
⚡	Simulation	bitwise_not_vortex_buffer_mut[2048]	398.6 ns	340.3 ns	+17.14%
⚡	Simulation	encode_varbin[(10000, 2)]	977.1 µs	835.6 µs	+16.94%
⚡	Simulation	chunked_varbinview_into_canonical[(100, 100)]	307.4 µs	272.1 µs	+12.96%
⚡	Simulation	encode_varbin[(1000, 512)]	210.7 µs	191 µs	+10.28%
🆕	Simulation	add_decimal_i128_nullable	N/A	6.8 ms	N/A
🆕	Simulation	add_decimal_i64_nonnull	N/A	3 ms	N/A

Tip

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

---

_{Comparing mk/decimal-add-sub (e3fdf45) with develop (d2b2378)}

Footnotes

1. 38 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. ↩