Canonicalize chunked nested types through the builder [builders-child-stack]

[robert3005]

Always use builders when canonicalizing chunked arrays

[codspeed-hq]

Merging this PR will regress 10 benchmarks

⚡ 30 improved benchmarks
❌ 10 regressed benchmarks
✅ 1894 untouched benchmarks
⏩ 51 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	canonicalize[1024, 32]	55.9 µs	82.7 µs	-32.35%
❌	Simulation	canonicalize[256, 32]	55.9 µs	81.8 µs	-31.65%
❌	Simulation	canonicalize[1024, 8]	44.1 µs	62.6 µs	-29.66%
❌	Simulation	canonicalize[16, 32]	58 µs	82.3 µs	-29.5%
❌	Simulation	canonicalize[256, 8]	44 µs	61.9 µs	-28.96%
❌	Simulation	canonicalize[1024, 2]	41.9 µs	57 µs	-26.54%
❌	Simulation	canonicalize[16, 8]	47.3 µs	61.8 µs	-23.46%
❌	Simulation	canonicalize[256, 2]	42.2 µs	54.5 µs	-22.61%
❌	Simulation	canonicalize[16, 2]	53.7 µs	65.1 µs	-17.49%
❌	Simulation	chunked_dict_primitive_into_canonical[u32, (1000, 10, 100)]	1.6 ms	1.8 ms	-10.42%
⚡	Simulation	take[small_m/shuffled/primitive/nonnull/chunks=16384/indices=1]	1,079.2 µs	74.4 µs	×15
⚡	Simulation	take[value_shape/duplicate99/fsl8/nonnull/chunks=16/indices=10000]	36.6 ms	10.6 ms	×3.5
⚡	Simulation	extend_from_array_zctl[(1000, 64)]	2,187.2 µs	656.4 µs	×3.3
⚡	Simulation	take[value_shape/shuffled/fsl8/nonnull/chunks=16/indices=10000]	37.6 ms	11.8 ms	×3.2
⚡	Simulation	extend_from_array_zctl[(10000, 8)]	3.6 ms	1.3 ms	×2.9
⚡	Simulation	take_chunked_fsl_random[32, 64]	1,110.7 µs	514 µs	×2.2
⚡	Simulation	take[small_m/shuffled/primitive/nonnull/chunks=1024/indices=1]	156.2 µs	74.1 µs	×2.1
⚡	Simulation	take[core/sorted/primitive/nonnull/chunks=16/indices=100000]	29.8 ms	14.3 ms	×2.1
⚡	Simulation	fsl_large	200.3 µs	106.6 µs	+87.89%
⚡	Simulation	take_chunked_fsl_random[16, 64]	863.5 µs	469.9 µs	+83.76%
...	...	...	...	...	...

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 claude/chunked-canonical-via-builder-9ze0t6 (5f173bf) with claude/builders-bulk-loop-callers-9ze0t6 (64c3464)}

Footnotes

1. 51 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. ↩