feat: add GpuIndexHNSW — GPU HNSW search with OCQ beam search kernel (slim, feature-only)

[devin-ai-integration]

Summary

Slim, review-friendly version of the GPU HNSW feature currently on gpu-hnsw-faiss (PR #1).

The gpu-hnsw-faiss branch mixes the feature with a full upstream faiss OSS sync (via a Merge upstream/main commit), so its diff-vs-main is ~9.5k lines / 194 files. This branch rebases only the feature onto current main and squashes the noisy development history (perf experiments + their reverts) into one commit, so the diff is exactly the feature:

faiss/gpu/GpuIndexHNSW.{cu,h}
faiss/gpu/impl/GpuHnsw{Build,Search,SearchKernel}.cuh
faiss/gpu/impl/GpuHnswTypes.{cu,h}
faiss/gpu/CMakeLists.txt
→ 8 files, +1898

No upstream sync, no reverted char4/dp4a experiments — just the final feature.

Relationship to other PRs:

• Supersedes the reviewability of feat: GPU HNSW search with parallel beam-search kernel — native FP32/FP16/BF16/INT8 #1 (gpu-hnsw-faiss → main). If we prefer to keep landing via gpu-hnsw-faiss, this branch is still useful as the clean diff to review against.
• The dead-code removal in refactor: remove dead cooperative-distance code from GPU HNSW kernel #2 (cooperative-distance block) is not included here yet — this branch is a faithful slim of the current feature; refactor: remove dead cooperative-distance code from GPU HNSW kernel #2 can be applied on top or folded in on request.

Not build-verified locally (no CUDA toolchain wired for a full build in this session); tree is byte-identical to gpu-hnsw-faiss^1.

Link to Devin session: https://6sense.devinenterprise.com/sessions/d39aba56b8a3467cbbf231ab631a06ed

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

Superseded: this content is now folded directly into the clean gpu-hnsw-faiss branch (rebased onto synced upstream master, dead-code/GPU_TIMING cleanups included). Closing.