refactor(array): give canonical decimals one integer child

[connortsui20]

Stacked on #10333. This experimental draft remains at the top of the stack.

A decimal with precision 76 can contain values that fit in 32 bits. Giving its canonical representation a plain 256-bit integer buffer would expand the value buffer eightfold before any computation.

Decimal precision now determines the logical dtype of one signed integer child. Narrow retains smaller stored values and encoded children during canonicalization, so selection, comparison, and supported aggregates can work through the child. Consumers that need native buffers explicitly materialize values at the stored width.

The experiment adds signed 128-bit and 256-bit integer extensions and retains the existing Decimal wire representation.

[codspeed]

Merging this PR will regress 3 benchmarks

⚠️ 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

⚡ 21 improved benchmarks
❌ 3 regressed benchmarks
✅ 2122 untouched benchmarks
⏩ 476 skipped benchmarks¹
🗄️ 1 archived benchmark run²

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	random_i128[0.01]	9.6 µs	23.1 µs	-58.38%
❌	Simulation	patterns_i128[Contiguous]	20.2 µs	30.2 µs	-33.07%
❌	Simulation	in_place_non_nullable[16384]	37.9 µs	49.3 µs	-23.06%
⚡	Simulation	random_i128[0.8]	177.3 µs	22.5 µs	×7.9
⚡	Simulation	random_i128[0.95]	161.9 µs	22.6 µs	×7.2
⚡	Simulation	random_i256[0.95]	127 µs	22.6 µs	×5.6
⚡	Simulation	random_i256[0.8]	125.8 µs	22.6 µs	×5.6
⚡	Simulation	random_i256[0.5]	122.1 µs	22.5 µs	×5.4
⚡	Simulation	patterns_i128[Random]	119.2 µs	22.5 µs	×5.3
⚡	Simulation	random_i128[0.5]	119.2 µs	22.6 µs	×5.3
⚡	Simulation	patterns_i128[Runs]	119.8 µs	23.6 µs	×5.1
⚡	Simulation	cached_indices_i128[0.1]	41.2 µs	21.1 µs	+95.8%
⚡	Simulation	random_i256[0.01]	39.2 µs	23.2 µs	+68.74%
⚡	Simulation	decompress[u32, (1000, 256)]	54.9 µs	32.9 µs	+66.97%
⚡	Simulation	cached_indices_i128[0.01]	46.5 µs	30.8 µs	+50.64%
⚡	Simulation	cached_indices_i32[0.1]	102.6 µs	78.5 µs	+30.68%
⚡	Simulation	take_map[(0.1, 1.0)]	289.3 µs	221.7 µs	+30.52%
⚡	Simulation	in_place_nullable[16384]	32.7 µs	26.3 µs	+24.53%
⚡	Simulation	random_i64[0.8]	167 µs	136.9 µs	+21.94%
⚡	Simulation	random_i64[0.95]	174.4 µs	143.9 µs	+21.21%
...	...	...	...	...	...

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/decimal-integer-child (173be79) with ct/narrow-compression (71f9458)}

Footnotes

1. 476 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. 1 benchmark was run, but is now archived. If it was deleted in another branch, consider rebasing to remove it from the report. Instead if it was added back, click here to restore it. ↩