fix: IVF nearest-partition basic test returns an unexpected class ID

[vsag-bot]

Summary

Corrected the flaky IVF basic test by comparing routing results produced with identical HGraph search breadth.

Root Cause

ClassifyDatas used ef_search=10 while the test's direct approximate HGraph search used ef_search=20. Different search breadths can legitimately select different partition IDs, making the assertion intermittent.

Changes

• Set the basic test reference search to ef_search=10.
• Documented that the comparison validates IVF routing rather than HGraph search breadth.

Validation

• cmake --build build --parallel 4: passed - Generic-SIMD build completed with DISABLE_SSE_FORCE=1.
• ./build/tests/unittests 'IVF Nearest Partition Basic Test' --reporter compact (50 repetitions): passed - All repetitions passed, exercising threaded and serial paths.
• ./build/tests/unittests '[IVFNearestPartition]' --reporter compact: passed - 3003 assertions passed across both IVF nearest-partition tests.
• clang-format-15 --dry-run --Werror src/algorithm/ivf/ivf_nearest_partition_test.cpp: passed - Validated with clang-format 15.0.7.
• git diff --check: passed - No whitespace errors.

Residual Risks

• The original intermittent mismatch did not reproduce in 50 local baseline runs; its mechanism is established from the differing search parameters and CI evidence.
• The ASan daily-job profile was not rerun locally.
• During initial dependency setup, CMake fetched a public ANTLR archive from GitHub before that build was stopped; no credentials or repository APIs were used.

Base branch: main

Fixes #2594

Prepared by vsag-bot after automatic effort ordering and AI repairability preflight.

[vsag-bot]

/label status/waiting-for-review
/waiting-on reviewer
/request-review @jiaweizone
/request-review @wxyucs
/request-review @inabao

[vsag-bot]

AI PR state: status/waiting-for-review
Current owner: @wxyucs
Reason: 1 review thread(s) remain unresolved
Head: 70d528d

Automatic merging is disabled. A maintainer performs the final merge.

[mergify]

Merge Protections

🟢 All 3 merge protections satisfied — ready to merge.

Show 3 satisfied protections

🟢 Require kind label

• label~=^kind/

🟢 Require version label

• label~=^version/

🟢 Require linked issue for feature/bug PRs

• body~=(?im)(?:^|[\s\-\*])(?:close[sd]?|fix(?:e[sd])?|resolve[sd]?)\s*:?\s+(?:#\d+|[\w.\-]+/[\w.\-]+#\d+|https?://github\.com/[\w.\-]+/[\w.\-]+/issues/\d+)

[LHT129]

[note] The fix correctly aligns the test's ef_search with ClassifyDatas' internal value (max(10, buckets_per_data*1.2) = 10 when buckets_per_data=1). Different HGraph search breadths can legitimately produce different nearest neighbors, so the previous ef_search=20 was testing HGraph search quality rather than IVF routing correctness.

The Serialize Test (line 99) still uses ef_search=20, which is fine — that test validates serialization round-trip behavior, not routing equivalence.

LGTM for this change.