Add Zeelin UIC-183 for angular benchmarks

[thu-nmrc]

Add Zeelin UIC-183 for angular ANN benchmarks

Summary

This PR adds Zeelin UIC-183, an angular/cosine approximate nearest-neighbor implementation for ANN-Benchmarks.

The submitted algorithm is exposed as:

zeelin-uic183

The implementation class is:

ZeelinLargecapPrefixbudgetM10BoundarytrimHnswgateIvf

The primary submission target is the default ANN-Benchmarks single-query leaderboard for angular, k=10.

Files Added

• ann_benchmarks/algorithms/zeelin_uic183/module.py
• ann_benchmarks/algorithms/zeelin_uic183/config.yml
• ann_benchmarks/algorithms/zeelin_uic183/Dockerfile
• ann_benchmarks/algorithms/zeelin_uic183/README.md

The CI benchmark matrix is updated to include zeelin_uic183.

Reproduction

python3 install.py --algorithm zeelin_uic183
python3 run.py --dataset glove-100-angular --algorithm zeelin-uic183 --runs 9
python3 run.py --dataset glove-25-angular --algorithm zeelin-uic183 --runs 9
python3 run.py --dataset nytimes-256-angular --algorithm zeelin-uic183 --runs 9

Local Official-Runner Results

These numbers were computed from ANN-Benchmarks runner.run() HDF5 outputs using the official knn and QPS metric functions.

Dataset	Recall	QPS	Public frontier QPS	Ratio
glove-100-angular	0.10024	160733	51650	3.11x
glove-25-angular	0.16277	212323	67354	3.15x
nytimes-256-angular	0.11473	112771	55231	2.04x

Notes

The algorithm also supports batch mode, but this PR is intended for the default single-query benchmark.