[MOD-16696] Use one-byte HNSW node locks

[GuyAv46]

Summary

Revives the old one-byte mutex idea on top of current main, but keeps the graph-data layout natural by moving per-node HNSW locks into an index-owned lock vector.

Changes:

• Adds vecsim_stl::one_byte_mutex, backed by std::atomic<uint8_t>::wait/notify_one.
• Removes std::mutex neighborsGuard from ElementGraphData.
• Adds elementLocks, a per-internal-id lock vector owned by HNSWIndex and resized with index capacity.
• Updates HNSW lock sites to lock by internal id instead of graph-data pointer.
• Updates HNSW memory estimation and allocator test accounting for the new 1-byte-per-node lock vector.

Expected memory effect:

• ElementGraphData no longer carries a std::mutex per vector.
• The replacement storage is one byte per capacity slot, in a separate vector.
• This avoids extending packed struct layout over graph-data pointers/flexible link arrays.

Validation

Local validation in a fresh worktree from origin/main:

• make build
• ctest --output-on-failure -R 'test_allocator|test_hnsw|test_hnsw_parallel' -j$(nproc)
• ./test_hnsw_parallel
• ./test_allocator --gtest_filter='*hnsw*'
• ROOT=/home/guy/Code/VectorSimilarity-hnsw-one-byte-locks ./test_hnsw --gtest_brief=1

Note: make check-format currently reports pre-existing unrelated formatting violations in files outside this change (for example batch_iterator.h, benchmark headers, and existing constructors in test_allocator.cpp). The files touched in this PR were formatted selectively.

Benchmark request

Please run benchmark comparison for this PR. I am adding the benchmarks-all label to trigger the benchmark workflow.

---

Note

Medium Risk
Touches core HNSW concurrency on graph mutations and parallel search; correctness depends on the new lock primitive and id-indexed locking staying aligned with existing deadlock ordering.

Overview
HNSW neighbor-link locking moves out of packed graph nodes into an index-owned elementLocks vector keyed by internal id, with a new compact vecsim_stl::one_byte_mutex (atomic wait/notify) instead of std::mutex on each ElementGraphData.

lockNodeLinks / unlockNodeLinks now take only idType; search, insert, delete/repair, batch iteration, and debug paths were updated accordingly. elementLocks is initialized on construct/load and grows with resizeIndexCommon. Memory estimation and allocator tests count one byte per capacity slot for the lock vector.

^{Reviewed by Cursor Bugbot for commit e6da02b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[jit-ci]

🛡️ Jit Security Scan Results

✅ No security findings were detected in this PR

---

^{Security scan by Jit}

[codecov]

Codecov Report

❌ Patch coverage is 96.22642% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 97.10%. Comparing base (45daaf3) to head (81dc0e6).

Files with missing lines	Patch %	Lines
src/VecSim/algorithms/hnsw/hnsw.h	94.73%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #983      +/-   ##
==========================================
- Coverage   97.12%   97.10%   -0.03%     
==========================================
  Files         141      141              
  Lines        8164     8176      +12     
==========================================
+ Hits         7929     7939      +10     
- Misses        235      237       +2     

☔ 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.

[cursor]

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

There are 3 total unresolved issues (including 2 from previous reviews).

^{❌ 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 81dc0e6. Configure here.}