Remove AlignedBoxWithSlice wrapper and add alias to Poly<[T], AlignedAllocator>

[JordanMaples]

Closes #907

This change introduces AlignedSlice<T> type alias and the aligned_slice() constructor in diskann-quantization::alloc, replacing the AlignedBoxWithSlice wrapper struct from diskann-providers. This eliminates an unnecessary abstraction layer and moves aligned allocation to its proper tier.

• Add AlignedSlice<T> = Poly<[T], AlignedAllocator> type alias
• Add aligned_slice<T>(capacity, alignment) constructor
• Migrate all 9 consumer files in diskann-disk
• Replace split_into_nonoverlapping_mut_slices with chunks_mut
• Replace as_slice/as_mut_slice with Deref/DerefMut
• Delete aligned_allocator.rs from diskann-providers
• Add 5 tests for aligned_slice

[JordanMaples]

This replaces #908

[Copilot]

Pull request overview

This PR removes the AlignedBoxWithSlice wrapper from diskann-providers and standardizes aligned slice allocation in diskann-quantization::alloc via a new AlignedSlice<T> alias and aligned_slice() constructor, updating disk-index readers/search code to use the new API.

Changes:

• Add AlignedSlice<T> = Poly<[T], AlignedAllocator> and aligned_slice<T>(capacity, alignment: PowerOfTwo) in diskann-quantization.
• Migrate aligned buffer usage across diskann-disk (tests/benches/search) to the new constructor and to slice deref semantics.
• Remove diskann-providers::common::AlignedBoxWithSlice and replace custom splitting with chunks_mut.

Reviewed changes

Copilot reviewed 15 out of 15 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
diskann-quantization/src/alloc/mod.rs	Exposes new aligned-slice module + re-exports AlignedSlice/aligned_slice.
diskann-quantization/src/alloc/aligned_slice.rs	Introduces AlignedSlice<T> alias, constructor, and unit tests.
diskann-providers/src/common/mod.rs	Drops AlignedBoxWithSlice re-export.
diskann-providers/src/common/aligned_allocator.rs	Deletes the wrapper implementation (and its tests).
diskann-disk/src/utils/aligned_file_reader/windows_aligned_file_reader.rs	Updates tests to allocate aligned buffers via aligned_slice and split via chunks_mut.
diskann-disk/src/utils/aligned_file_reader/storage_provider_aligned_file_reader.rs	Same migration as above for storage-provider reader tests.
diskann-disk/src/utils/aligned_file_reader/linux_aligned_file_reader.rs	Same migration as above for Linux reader tests.
diskann-disk/src/utils/aligned_file_reader/aligned_read.rs	Updates/shortens doc comment reflecting removal of wrapper type.
diskann-disk/src/search/provider/disk_vertex_provider_factory.rs	Uses aligned_slice buffer for header reads and relies on deref slicing.
diskann-disk/src/search/provider/disk_sector_graph.rs	Switches internal storage to AlignedSlice<u8> and replaces custom splitting with chunks_mut.
diskann-disk/src/search/provider/disk_provider.rs	Removes .as_slice() calls; relies on deref to slice for .to_vec().
diskann-disk/src/search/pq/quantizer_preprocess.rs	Replaces .as_mut_slice() with deref-based mutable slicing/borrows.
diskann-disk/src/search/pq/pq_scratch.rs	Replaces wrapper fields with AlignedSlice allocations and updates alignment tests.
diskann-disk/benches/benchmarks_iai/aligned_file_reader_bench_iai.rs	Updates benchmark to use aligned_slice + chunks_mut.
diskann-disk/benches/benchmarks/aligned_file_reader_bench.rs	Updates benchmark to use aligned_slice + chunks_mut.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 91.76471% with 7 lines in your changes missing coverage. Please review.
✅ Project coverage is 89.31%. Comparing base (f23d38b) to head (51104f2).

Files with missing lines	Patch %	Lines
...kann-disk/src/search/provider/disk_sector_graph.rs	71.42%	6 Missing ⚠️
diskann-disk/src/search/pq/quantizer_preprocess.rs	66.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #965      +/-   ##
==========================================
- Coverage   90.47%   89.31%   -1.17%     
==========================================
  Files         448      448              
  Lines       83447    83329     -118     
==========================================
- Hits        75497    74422    -1075     
- Misses       7950     8907     +957     

Flag	Coverage Δ
miri	89.31% <91.76%> (-1.17%)	⬇️
unittests	89.15% <91.76%> (-1.29%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
diskann-disk/src/search/pq/pq_scratch.rs	79.54% <100.00%> (+0.97%)	⬆️
diskann-disk/src/search/provider/disk_provider.rs	90.89% <ø> (-0.01%)	⬇️
...rc/search/provider/disk_vertex_provider_factory.rs	95.70% <100.00%> (+0.05%)	⬆️
...disk/src/utils/aligned_file_reader/aligned_read.rs	100.00% <ø> (ø)
...s/aligned_file_reader/linux_aligned_file_reader.rs	82.65% <100.00%> (-0.35%)	⬇️
...ile_reader/storage_provider_aligned_file_reader.rs	98.36% <100.00%> (-0.03%)	⬇️
diskann-quantization/src/alloc/aligned_slice.rs	100.00% <100.00%> (ø)
diskann-quantization/src/alloc/mod.rs	100.00% <ø> (ø)
diskann-disk/src/search/pq/quantizer_preprocess.rs	88.00% <66.66%> (ø)
...kann-disk/src/search/provider/disk_sector_graph.rs	96.94% <71.42%> (-1.24%)	⬇️

... and 39 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[hildebrandmw]

Thanks for the cleanup @JordanMaples! This is a step in the right direction, but I have some concrete requests for a follow-up. While this PR removes AlignedBoxWithSlice to the letter, it doesn't remove it in spirit as the AlignedSlice alias and aligned_slice constructor still exist. We should strive for construction of concrete types to not require a helper free function.

Instead, I would like call sites to explicitly compose the container and allocator instead of using the thin helper. We can make this more ergonomic by introducing const aliases for all valid PowerOfTwo values:

impl PowerOfTwo {
    // Powers
    const P0: Self = unsafe { Self::new_unchecked(NonZeroUsize::new_unchecked(1)) };
    const P1: Self = unsafe { Self::new_unchecked(NonZeroUsize::new_unchecked(2)) };
    // ...
    const P63: Self = unsafe { Self::new_unchecked(NonZeroUsize::new_unchecked(1 << 63)) };

    // Values
    const V1: Self = Self::P0;
    const V2: Self = Self::P1;
    // ...
    const V9223372036854775808: Self = Self::P63;
}

Then in situations where the alignment is known and static, we can replace

// Before
let buf = aligned_slice::<u8>(sector_len, PowerOfTwo::new(512)?)?;

// After
let buf = Poly::broadcast(
    0u8,
    sector_len, 
    AlignedAllocator::new(PowerOfTwo::V512)
)?;

We can also add const aliases for AlignedAllocator for common alignments to simplify even further:

let buf = Poly::broadcast(0u8, sector_len, AlignedAllocator::A512)?;

This way, both the initialization of the allocator and the data are visible at the callsite.

My concrete ask is this:

1. Add PowerOfTwo/AlignedAllocator const aliases.
2. Delete aligned_slice.rs and update everything to use Poly directly.

[hildebrandmw]

Note: we probably never need to ever materializes the slices as a Vec<&mut [u8]>. In the context of the benchmark, a chunks_mut iterator would have worked just fine. In other situations, par_chunks directly would work as well.