refactor: make bitpacked CPU kernels consume stored chunk layouts

[mhk197]

Make CPU readers locate each 1024-value chunk through the offsets child and derive its width as (end - start) / 128. This replaces the assumption that every chunk has one scalar width and a fixed byte stride.

Bulk decoding, take, filter, comparisons, streaming predicates, constant detection, and fused frame-of-reference decoding prepare and validate ChunkLayout once per operation. They borrow materialized offsets or execute the compressed child once, then use direct buffer access in their loops. Scalar decoding reads only the selected chunk's boundaries, using execute_scalar when necessary. Slicing preserves the relevant boundaries and their origin, including partial and zero-width chunks.

Remove the temporary scalar bit_width and allow compressed offsets. Validate boundary differences and packed-buffer bounds before unpacking, with regression tests for malformed materialized and compressed children. Encoders still choose a uniform width, and serialization still accepts only the original v1 format. CUDA retains its uniform-width restriction.

Validation: 396 FastLanes/BtrBlocks tests passed (1 skipped). Focused Clippy passed with all targets, all features, and warnings denied.

[codspeed]

Merging this PR will regress 1 benchmark

⚠️ 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.

⚠️ 4 benchmarks measured no execution time

Nothing ran under measurement, usually because the compiler removed the code under test. These results are not comparable, so they count as unchanged.

Preventing compiler optimizations

⚠️ 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

⚡ 2 improved benchmarks
❌ 1 regressed benchmark
✅ 2229 untouched benchmarks
⏩ 329 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	take_fsl_random[128, 10]	32.7 µs	58.5 µs	-44.13%
⚡	Simulation	take_fsl_u8_random[256, 100]	98.7 µs	45.8 µs	×2.2
⚡	Simulation	take_fsl_nullable_random[16, 100]	96.9 µs	50.5 µs	+92.04%
⚠️	Simulation	take_fsl_f16_random[16, 100]	61.5 µs	< 1 ns	N/A
⚠️	Simulation	fixed_16_advancing_ptr_safe[100]	< 1 ns	< 1 ns	N/A
⚠️	Simulation	preverify_advancing_ptr_unchecked[1000]	< 1 ns	< 1 ns	N/A
⚠️	Simulation	preverify_advancing_ptr_unchecked[10000]	< 1 ns	< 1 ns	N/A

Tip

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

---

_{Comparing mk/bitpacked-stack-02-cpu-layout (797717f) with mk/bitpacked-stack-04-offset-child (06905e1)}

Footnotes

1. 329 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. ↩