Fix decimal scalar Mul/Div to rescale at the shared scale

[mhk197]

What

DecimalScalar::checked_binary_numeric applied the checked operator to the raw stored integers with no scale handling, so Mul results were off by 10^scale (its own test asserted 0.50 × 0.10 = "5.00"; mathematically 0.05) and Div dropped the scale entirely (10.00 / 0.10 = "1.00" instead of 100.00).

This fixes Mul and Div to fixed-point semantics at the shared scale, truncating toward zero:

• Mul: the raw product (scale 2s) is rescaled back to s via a new truncating rescale helper. The product is computed in i256, so it can exceed the operand storage width as long as the rescaled result fits the precision.
• Div: (a · 10^s) / b for s ≥ 0, a / (b · 10^{-s}) for negative scales; division by zero returns None.
• Add/Sub are unchanged (raw ops are exact at a shared scale).

The raw DecimalValue::checked_mul/checked_div remain scale-oblivious and are now documented as such, pointing at the new checked_mul_fixed_point/checked_div_fixed_point.

The only production callers of the scalar op use Add (sum aggregate, stats), so this is a safe bug fix. It is also a prerequisite for native decimal array arithmetic (follow-up PR), whose conformance tests compare element-wise against this scalar implementation.

Note: the truncating rescale helper takes i16 scales because intermediate scales (e.g. 2s for the raw product) can exceed the i8 range.

Testing

• Fixed the Mul/Div test expectations and added truncation (0.15 × 0.15 → 0.02, toward-zero for negatives), negative-scale, wider-than-storage, and precision-overflow cases.
• cargo nextest run -p vortex-array, cargo clippy --all-targets --all-features, cargo +nightly fmt --all.

🤖 Generated with Claude Code

[codspeed-hq]

Merging this PR will not alter performance

⚡ 6 improved benchmarks
❌ 2 regressed benchmarks
✅ 1629 untouched benchmarks
⏩ 42 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	chunked_varbinview_canonical_into[(1000, 10)]	154.6 µs	190.9 µs	-19.02%
❌	Simulation	slice_empty_vortex	339.4 ns	397.8 ns	-14.66%
⚡	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	+16.93%
⚡	Simulation	chunked_varbinview_opt_canonical_into[(100, 100)]	340.6 µs	305.7 µs	+11.41%
⚡	Simulation	compare_int_constant	298.7 µs	268.4 µs	+11.28%

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-scalar-fixed-point (591320d) with develop (d2b2378)}

Footnotes

1. 42 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. ↩

[mhk197]

Superseded by the per-operator restructure starting at #8724. The scalar Mul and Div fixes from this PR will land with the decimal Mul and Div kernel PRs respectively, since each fix is only observable through the op it corrects.