Implement SVSIndex::getDataByLabel (MOD-17706)

[nonirosenfeldredis]

Describe the changes in the pull request

SVSIndex::getDataByLabel was a stub that always reported nothing where VecSimTieredIndex::getDataByLabel had a bypass for SVS backend.

It now reads the stored vectors directly from the underlying SVS index (single-value via
get_datum, multi-value via get_label_to_external_lookup + get_parent_index), refusing only for
compressed/LeanVec-reduced storage (storage_traits_t::is_compressed()) -- where it reports nothing.

The read logic is shared with the existing test-only getStoredVectorDataByLabel via a new private
appendStoredDataByLabel helper, so the single/multi-value branching and the raw memcpy live in one
place.

Tiered read logic

VecSimTieredIndex::getDataByLabel always reads the flat buffer first, then decides whether the
backend also needs reading: for a single-value label the buffer alone is the whole answer once it
has contributed anything, but a multi-value label's vectors are routinely split across the tiers
while an ingest job is pending, so the backend is always consulted too in that case.

A compressed SVS backend complicates that decision: it appends nothing from getDataByLabel
regardless of whether it holds part of the label -- it cannot report values, which is not the
same as "doesn't have any". Unlike HNSW, whose tiered backend can never be quantized, SVS's
routinely is, so a multi-value label split between a non-empty buffer contribution and such a
backend would otherwise come back as just the buffer's subset, indistinguishable from the complete
answer. Before trusting that subset, the tiered read checks backend membership via
SVSIndexBase::isLabelExists -- cheap because it's metadata, not the value read compression rules
out -- and if the backend is compressed and holds part of the label, discards the buffer's
contribution too and reports nothing for the whole label, the same rule SVSIndex::getDataByLabel
already applies within a single tier.

Which issues this PR fixes

1. MOD-17706

Main objects this PR modified

1. src/VecSim/algorithms/svs/svs.h
2. src/VecSim/vec_sim_tiered_index.h
3. tests/unit/test_svs.cpp
4. tests/unit/test_svs_tiered.cpp

Mark if applicable

• This PR introduces API changes
• This PR introduces serialization changes

---

Note

Medium Risk
Changes observable getDataByLabel behavior for SVS and tiered SVS (including RediSearch “holds vectors” paths); logic for compressed backends and multi-value labels split across tiers is easy to get wrong.

Overview
Implements SVSIndex::getDataByLabel so uncompressed SVS indexes return the stored vectors for a label (single-value via get_datum, multi-value via label lookup + parent index). Compressed or LeanVec-reduced indexes still append nothing, matching HNSW—no dequantized bytes that could fool equality checks.

The read path is centralized in a new appendStoredDataByLabel helper, also used by the test-only getStoredVectorDataByLabel, with compression gated on storage_traits_t::is_compressed() instead of separate quant flags.

Tiered getDataByLabel no longer skips all backend reads for SVS. It always consults the flat buffer first, then the backend when needed (multi-value or label not fully in the buffer). For a compressed SVS backend, if the buffer contributed vectors but the backend also holds that label (isLabelExists), the tiered read drops the buffer result and returns nothing so a partial subset is not mistaken for the full label.

Unit tests cover uncompressed vs compressed, single vs multi, absent labels, tiered flat-only reads, tiered backend reads after ingest, and the compressed split-label case.

^{Reviewed by Cursor Bugbot for commit ee9745e. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Stale Bugbot comment from a previous run.

[codecov]

Codecov Report

❌ Patch coverage is 97.05882% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 97.40%. Comparing base (08b33bf) to head (ee9745e).

Files with missing lines	Patch %	Lines
src/VecSim/vec_sim_tiered_index.h	92.85%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1033      +/-   ##
==========================================
- Coverage   97.41%   97.40%   -0.01%     
==========================================
  Files         141      141              
  Lines        8688     8698      +10     
==========================================
+ Hits         8463     8472       +9     
- Misses        225      226       +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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

[dor-forer]

For a compressed SVS backend with no buffer contribution, this acquires mainIndexGuard even though the result must be empty. It can wait behind a batch’s exclusive addVectors lock, then immediately return nothing from SVSIndex::getDataByLabel. While waiting, it also holds flatIndexGuard shared, potentially delaying buffer writes.
Could we resolve svs_backend before acquiring the main lock and return early when isCompressed() && vectors_output.size() == before_flat?
The multi-value case with a buffer contribution should retain the locked isLabelExists check, since we still need to determine whether the buffer represents the complete label.

[nonirosenfeldredis]

done

[dor-forer]

Consider adding a same type optimization (if OutputElement and DataType),
We can create vec_data directly in vectors_output.
Something like:

auto append_datum = [&](auto indexed_span) {
    if constexpr (std::is_same_v<OutputElement, DataType>) {
        vectors_output.emplace_back(indexed_span.begin(), indexed_span.end());
    } else {
        // Existing implementation, preserved for FP16 and byte output.
        std::vector<OutputElement> vec_data(
            this->getStoredDataSize() / sizeof(OutputElement));
        std::memcpy(vec_data.data(), indexed_span.data(), this->getStoredDataSize());
        vectors_output.push_back(std::move(vec_data));
    }
};

[nonirosenfeldredis]

done

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit fa5b616. Configure here.}