fix(bool): trim oversized backing buffer views

[jiaqizho]

Rationale for this change

Bool arrays constructed from sliced bitmaps can retain buffer views that extend beyond their logical byte range. Serialization writes the exposed buffers in full, which can unnecessarily increase file size and retain logically excluded suffix data.

This affects the release path of BoolArray::new_unchecked and callers that provide an oversized BufferHandle.

What changes are included in this PR?

Normalize unchecked host bit buffers with shrink_offset, and trim oversized host or device buffer handles to the byte range required by the Bool metadata.

Both operations create zero-copy views and preserve the underlying allocation, device residency, alignment, and ownership model.

What APIs are changed? Are there any user-facing changes?

No public API signatures are changed. Bool array constructors now expose only the backing bytes required by their logical offset and length.

[joseph-isaacs]

If this problem does exist it would be great to fix this. Does it show up in a serialised file (on disk)?

[codspeed-hq]

Merging this PR will degrade performance by 21.19%

❌ 1 regressed benchmark
✅ 1846 untouched benchmarks
⏩ 46 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	decode_varbin[(1000, 8)]	27.1 µs	34.4 µs	-21.19%

Tip

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

---

_{Comparing jiaqizho:fix-oversized-bool-array (3a8c077) with develop (30091f2)}

Footnotes

1. 46 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. ↩

[jiaqizho]

If this problem does exist it would be great to fix this. Does it show up in a serialised file (on disk)?

Hi @joseph-isaacs

For context, I originally encountered a similar issue while using a custom layout rather than Vortex's default layout.

I reproduced the issue on master using the following minimal writer strategy. The direct FlatLayoutStrategy setup is intended only to isolate the underlying behavior from my custom layout:

  session
      .write_options()
      .with_strategy(Arc::new(FlatLayoutStrategy::default()))
      .write(&mut file, array.to_array_stream())
      .await?;

The input contains 8,192 nullable Boolean rows. The logical values and validity bitmaps are 1,024 bytes each, but each is backed by a 65,536-byte BufferHandle.

Results:

• PR base: 132,828-byte file
• with this patch: 3,804-byte file
• Generated files: direct-flat-vortex.zip