feat(array): add Narrow integer encoding

[connortsui20]

Summary

Progress towards #10020. Related prototype: #10015.

Integer arrays can retain their logical dtype while storing a narrower child, so selecting rows and comparing values can use the smaller representation. This is also the first step towards giving canonical decimals an integer child whose logical width follows the decimal precision.

Changes

NarrowArray preserves signedness and nullability, accepts encoded children, and widens through a lazy cast when canonical primitive values are needed. Selection, compatible casts, fitting fill values, comparisons, and integer aggregates operate through the child. Aggregate partial states are converted back to the logical dtype, and arithmetic keeps the logical overflow behavior through canonical fallback.

This draft leaves compressor selection and decimal integration for follow-up work.

#10313 discusses whether dtype and range shortcuts may skip value-dependent errors from lazy children.

[codspeed]

Merging this PR will improve performance by 26.7%

⚠️ 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.

⚡ 29 improved benchmarks
✅ 2072 untouched benchmarks
🆕 3 new benchmarks
⏩ 518 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	fsst_decompress_string	526 µs	188.4 µs	×2.8
⚡	Simulation	chunked_canonicalize_into[(20, 100, 16, 4)]	739.9 µs	340.4 µs	×2.2
⚡	Simulation	chunked_dict_fsst_into_canonical[(250, 10, 2)]	681.9 µs	342.6 µs	+99.03%
⚡	Simulation	decode_primitives[i64, (1000, 32)]	39.9 µs	26.1 µs	+53.1%
⚡	Simulation	chunked_canonicalize_into[(20, 50, 8, 4)]	274.4 µs	213 µs	+28.85%
⚡	Simulation	chunked_canonicalize_into[(20, 50, 8, 16)]	279.5 µs	217.7 µs	+28.4%
⚡	Simulation	chunked_canonicalize_into[(20, 50, 8, 64)]	306.3 µs	242.6 µs	+26.29%
⚡	Simulation	chunked_into_canonical[(20, 50, 8, 4)]	314.3 µs	252.2 µs	+24.58%
⚡	Simulation	chunked_into_canonical[(20, 50, 8, 16)]	323.3 µs	261 µs	+23.87%
⚡	Simulation	chunked_into_canonical[(20, 50, 8, 64)]	348.2 µs	285.6 µs	+21.91%
⚡	Simulation	decompress_fsst[(200, 4, 4)]	60.8 µs	50.6 µs	+20.03%
⚡	Simulation	decompress_fsst[(200, 4, 8)]	61.6 µs	51.6 µs	+19.45%
⚡	Simulation	chunked_canonicalize_into[(10, 150, 4, 4)]	235.3 µs	198.8 µs	+18.34%
⚡	Simulation	decompress_fsst[(200, 16, 4)]	66.4 µs	56.7 µs	+17.14%
⚡	Simulation	decompress_fsst[(200, 16, 8)]	69.4 µs	59.3 µs	+16.97%
⚡	Simulation	chunked_canonicalize_into[(10, 150, 16, 4)]	266.6 µs	230.1 µs	+15.88%
⚡	Simulation	chunked_canonicalize_into[(20, 100, 16, 16)]	461.3 µs	399.1 µs	+15.59%
⚡	Simulation	chunked_into_canonical[(10, 150, 4, 4)]	264.9 µs	230.3 µs	+15.03%
⚡	Simulation	chunked_into_canonical[(10, 150, 16, 4)]	298.7 µs	262.4 µs	+13.82%
⚡	Simulation	chunked_into_canonical[(20, 100, 16, 4)]	452.7 µs	397.9 µs	+13.78%
...	...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

Tip

Curious why performance improved? Comment @codspeedbot explain why performance improved on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing ct/narrow-encoding (44780e8) with develop (3c87da0)²}

Footnotes

1. 518 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. No successful run was found on develop (cc018ee) during the generation of this report, so 3c87da0 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩