feat(compress): retain Narrow integer children through compression

[connortsui20]

Summary

Progress towards #10020. Stacked on #10332.

Compressing a narrow integer child lets codecs decode at the smaller width while a Narrow wrapper retains the column's logical type. Session presets now use this representation when preview2026.10.0 is enabled. Core-only and CUDA presets keep it disabled.

Changes

The old primitive narrow method is replaced by shared width selection in NarrowArray. Internal buffers whose dtypes are chosen at construction use encode_values; arrays with a logical dtype contract use encode. Both retain signedness, so nonnegative signed internal buffers no longer silently become unsigned.

API Changes

Replaces PrimitiveArrayExt::narrow with NarrowArray::encode_values for internal buffers. Use NarrowArray::encode when the logical dtype must be retained.

[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

⚡ 10 improved benchmarks
❌ 3 regressed benchmarks
✅ 2091 untouched benchmarks
⏩ 518 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	take_map[(0.1, 1.0)]	225.3 µs	289.3 µs	-22.12%
❌	WallTime	bitpack_blocked_compress_avx2	6.7 µs	7.6 µs	-11.67%
❌	Simulation	take_map[(0.1, 0.5)]	147.7 µs	166.8 µs	-11.44%
⚡	Simulation	encode_primitives[u8, (1000, 2)]	117.3 µs	74 µs	+58.52%
⚡	Simulation	encode_primitives[u8, (1000, 4)]	117.5 µs	74.2 µs	+58.36%
⚡	Simulation	encode_primitives[u8, (1000, 8)]	118.7 µs	75.3 µs	+57.72%
⚡	Simulation	encode_primitives[u8, (1000, 32)]	120.1 µs	77 µs	+55.97%
⚡	Simulation	encode_primitives[u8, (1000, 512)]	136.1 µs	92.8 µs	+46.68%
⚡	Simulation	encode_primitives[u8, (2000, 4)]	169.4 µs	125.9 µs	+34.54%
⚡	Simulation	encode_primitives[u8, (2000, 2)]	169.2 µs	125.8 µs	+34.48%
⚡	Simulation	encode_primitives[u8, (2000, 8)]	170.1 µs	126.9 µs	+34.09%
⚡	Simulation	encode_primitives[u8, (2000, 32)]	172.1 µs	128.8 µs	+33.6%
⚡	Simulation	encode_primitives[u8, (2000, 512)]	188.4 µs	145 µs	+29.91%

Tip

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

---

_{Comparing ct/narrow-compression (71f9458) with ct/narrow-pushdown (557a613)}

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. ↩