Improve rank fusion implementation in Anserini

[lintool]

Rank fusion is the process of combining two or more ranked lists to produce a better ranked result. This is also called "hybrid search" sometimes.

I need help improving rank fusion features in Anserini. This would be a good URA project.

Start with the following:

Building indexes from scratch requires downloading the huge parquet tarball, so let's use the prebuilt indexes instead.
This downloads the prebuilt indexes:

bin/run.sh io.anserini.search.SearchCollection -index beir-v1.0.0-robust04.flat -topics beir-robust04 -output runs/run.beir.flat.robust04.txt -bm25 -removeQuery
bin/run.sh io.anserini.search.SearchFlatDenseVectors -index beir-v1.0.0-robust04.bge-base-en-v1.5.flat -topics beir-robust04.bge-base-en-v1.5 -output runs/run.beir.bge-base-en-v1.5.flat.cached_q.robust04.txt -threads 16 -removeQuery

Copy them into Anserini's indexes/ folder:

cp -r ~/.cache/pyserini/indexes/lucene-inverted.beir-v1.0.0-robust04.flat.20221116.505594.d508fc770002a99a5dc3da3d0fa001b7/ ./indexes/lucene-inverted.beir-v1.0.0-robust04.flat
cp -r ~/.cache/pyserini/indexes/lucene-flat.beir-v1.0.0-robust04.bge-base-en-v1.5.20240618.6cf601.7750b4abbc60fe821c5948a81296f1d0/ ./indexes/lucene-flat.beir-v1.0.0-robust04.bge-base-en-v1.5

These regression scripts will now run:

python src/main/python/run_regression.py --verify --search --regression beir-v1.0.0-robust04.flat
python src/main/python/run_regression.py --verify --search --regression beir-v1.0.0-robust04.bge-base-en-v1.5.parquet.flat.cached

python src/main/python/run_fusion_regression.py --regression beir-v1.0.0-robust04

[lilyjge]

I'm working on it!