Share fixed-width structural array operations

[connortsui20]

Why

Primitive and Decimal are different logical types, but their structural operations both move fixed-width records from one values buffer and propagate validity. Keeping separate take/filter pipelines meant fixes and optimizations still had to be maintained twice.

What

• add a three-hook FixedWidthArray adapter for byte width, values access, and reconstruction
• run Primitive and Decimal through one take pipeline, including index normalization, null handling, validity propagation, piecewise ranges, and record dispatch
• retain the direct spare-capacity copy used by the previous Primitive piecewise kernel, now over shared byte records
• run both through one filter pipeline for values and validity
• retain AVX2 gathers for 4- and 8-byte records; share scalar paths for 1, 2, 16, 32, and uncommon widths
• preserve their public APIs, distinct VTables/array IDs, and serialized layouts

Review structure

1. Share one-buffer VTable plumbing.
2. Share the complete fixed-width take/filter machinery.

Issue provenance

• Existing and fixed: AVX2 accepted index == len and silently zeroed some larger invalid indices.
• Introduced during development and fixed: temporary Decimal API and arbitrary-dtype regressions, plus redundant per-range checks/capacity handling in the first shared piecewise kernel.

[codspeed-hq]

Merging this PR will degrade performance by 10.1%

❌ 1 regressed benchmark
✅ 1840 untouched benchmarks
⏩ 55 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	prebuilt_dense_lengths[1]	457.1 µs	508.4 µs	-10.1%

Tip

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

---

_{Comparing ct/fixed-width-primitive-decimal (dfc6a40) with develop (e58fb58)}

Footnotes

1. 55 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. ↩

[robert3005]

I thought there would be more lines removed, i.e. majority of Primitive and Decimal array should be gone?

[connortsui20]

The majority is gone, the reason there's extra code is all the boilerplate and extra files