DO NOT MERGE Combine PiecewiseSequence take stack

[danking]

Summary

Combines the non-benchmark PiecewiseSequence take stack into one PR:

• core PiecewiseSequence array
• FixedSizeList take using PiecewiseSequence
• run-take consumers for contiguous slices
• VarBin and VarBinView PiecewiseSequence take specializations
• cursor-copy slice materialization
• List take element gathering via PiecewiseSequence

Excludes the benchmark-only PR #8830.

Validation

• Ran all eight non-CUDA CodSpeed microbenchmark shards locally on this PR branch and on origin/develop using cargo codspeed build/run -m walltime.
• Parsed 1,759 common benchmark medians from both branches.
• At a 10% median walltime cutoff: 253 common benchmarks changed (161 faster, 92 slower), plus 128 PR-only benchmarks.
• Local walltime is noisier than CodSpeed simulation; the strongest intended improvements are in take_fsl and listview_rebuild.

[codspeed-hq]

Merging this PR will regress 1 benchmark

⚠️ 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

⚡ 32 improved benchmarks
❌ 1 regressed benchmark
✅ 1637 untouched benchmarks
🆕 128 new benchmarks
⏩ 53 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	varbinview_large	389.9 µs	446.4 µs	-12.66%
⚡	Simulation	struct_large	432.6 µs	38 µs	×11
⚡	Simulation	fsl_large	616.5 µs	87.8 µs	×7
⚡	Simulation	take_fsl_random[1024, 1000]	15.2 ms	6 ms	×2.5
⚡	Simulation	take_fsl_nullable_random[1024, 1000]	14.8 ms	6 ms	×2.5
⚡	Simulation	take_fsl_random[256, 1000]	3.5 ms	1.6 ms	×2.2
⚡	Simulation	take_fsl_nullable_random[256, 1000]	3.4 ms	1.6 ms	×2.1
⚡	Simulation	take_fsl_random[1024, 100]	1,568.1 µs	798 µs	+96.51%
⚡	Simulation	take_fsl_nullable_random[1024, 100]	1,535.9 µs	808.3 µs	+90.02%
⚡	Simulation	take_fsl_random[256, 100]	418.5 µs	237.7 µs	+76.06%
⚡	Simulation	take_fsl_random[64, 1000]	815.7 µs	468.1 µs	+74.28%
⚡	Simulation	take_fsl_nullable_random[64, 1000]	812 µs	482.8 µs	+68.19%
⚡	Simulation	take_fsl_nullable_random[256, 100]	415.8 µs	247.9 µs	+67.73%
⚡	Simulation	list_i32_large	172.3 µs	106.2 µs	+62.22%
⚡	Simulation	take_fsl_random[16, 1000]	237.9 µs	182.4 µs	+30.43%
⚡	Simulation	take_filter_list_slice_mask_random_indices[256, 50]	118.4 µs	93.2 µs	+27.06%
⚡	Simulation	take_filter_list_small_uncached_random_mask_random_indices[256, 10]	116.1 µs	91.6 µs	+26.82%
⚡	Simulation	take_filter_list_nullable_random_mask_random_indices[256, 50]	128.4 µs	101.3 µs	+26.78%
⚡	Simulation	take_filter_list_small_random_mask_random_indices[256, 10]	115.4 µs	91.1 µs	+26.69%
⚡	Simulation	take_filter_list_slice_mask_random_indices[768, 50]	118.6 µs	93.9 µs	+26.3%
...	...	...	...	...	...

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 codex/piecewise-sequence-combined-pr (012f18f) with develop (e6eeaae)}

Footnotes

1. 53 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. ↩

[github-actions]

Polar Signals Profiling Results

Latest Run

Status	Commit	Job	Attempt	Link
🟢 Done	012f18f		1	Explore Profiling Data

---

Powered by Polar Signals Cloud

[github-actions]

Benchmarks: Vortex queries 📖

Verdict: No clear signal (medium confidence)
Attributed Vortex impact: +4.4%
Engines: DataFusion No clear signal (+3.0%, low confidence) · DuckDB No clear signal (+5.7%, low confidence)
Vortex (geomean): 0.911x ➖
Parquet (geomean): 0.892x ✅
Shifts: Parquet (control) -10.8% · Median polish -9.5%

How to read Verdict and Engines

• Verdict: Overall PR-level signal after subtracting baseline drift estimated from Parquet control rows. It can be Likely improvement, Likely regression, or No clear signal.
• Engines: Per-engine attribution. DataFusion is compared against DataFusion/Parquet controls; DuckDB is compared against DuckDB/Parquet controls. This answers whether each engine improved or regressed independently.
• Confidence: Based on directional consistency, share of rows above the noise floor, and control-run noise.

---

datafusion / vortex-file-compressed (0.855x ✅, 2↑ 0↓)

name	PR 012f18f (ns)	base e6eeaae (ns)	ratio (PR/base)
vortex_q00/datafusion:vortex-file-compressed 🚀	9794835	11446321	0.86
vortex_q01/datafusion:vortex-file-compressed 🚀	6368989	7449028	0.86

datafusion / parquet (0.830x ✅, 2↑ 0↓)

name	PR 012f18f (ns)	base e6eeaae (ns)	ratio (PR/base)
vortex_q00/datafusion:parquet 🚀	19710916	23584739	0.84
vortex_q01/datafusion:parquet 🚀	4634462	5616200	0.83

duckdb / vortex-file-compressed (1.013x ➖, 0↑ 0↓)

name	PR 012f18f (ns)	base e6eeaae (ns)	ratio (PR/base)
vortex_q00/duckdb:vortex-file-compressed	10395704	10387695	1.00
vortex_q01/duckdb:vortex-file-compressed	6039147	5885688	1.03

duckdb / parquet (0.958x ➖, 0↑ 0↓)

name	PR 012f18f (ns)	base e6eeaae (ns)	ratio (PR/base)
vortex_q00/duckdb:parquet	23273908	24576114	0.95
vortex_q01/duckdb:parquet	9432237	9723548	0.97

---

No file size changes detected.