Major refactoring of indexing pipeline (IndexCollection, IndexHnswDenseVectors, and IndexInvertedDenseVectors),
extracting common code paths into AbstractIndexer.

+ Refactored to create HnswDenseSearcher and InvertedDenseSearcher; these will provide Python bindings (later)
+ Refactored SearchHnswDenseVectors and SearchInvertedDenseVectors as wrappers to provide main
+ Improved test coverage

* add PrebuiltIndexHandler
    * add a download progress bar
    * add MD5 checksum checking 
    * add gzip and unzip tarball functionalities
* add corresponding unittests

Also added HNSW int8 regressions: works for cosDPR-distill, issues remain with OpenAI Ada2

+ Improved alignment between SearchCollection and dense vector search classes.
+ Aligned ScoredDoc and ScoredDocs (was previously ScoredDocuments) as container objects for Lucene results.
+ Searchers now use ScoredDoc instead of class-specific Result objects.
+ Tweaked SearchCollection args to use proper camelCasing.
+ Consolidated BaseSearcher class for basic ranked list post-processing functionality.
+ Increased test coverage.

…ings (#2318)

Default of 2047 seems to be too aggressive for hops, getting OOM errors.

…and Sparse Representations Using Lucene" (#2317)

…_eval (#2330)

add pre-encoded topic bindings for pyserini 2CR
remove -optimize flag to reduce runtime

#2343)

…2346)

…2352)

)

+ Update cos-DPR HNSW indexes (+quantized index)
+ Add BGE HNSW indexes (+quantized index)
+ Add ability to specify topic enum

+ Added Enum topic bindings SPLADE++ ED and BGE for MS MARCO v1 passage corpus
+ Updated regression documentation
+ Rebalanced regression batches across batch03, batch04, and batch05

+ Condensed regression batches from three back into two.
+ Adjusted score tolerance to 0.007 for HNSW runs.

- updated splade-pp-ed-optimized.onnx on remote end
- updated onnxruntime to support opset 19

Adds 2CR regressions doc for MS MARCO embedded with cohere embed-english-v3

Added int8 indexes, general clean-up, added links to README

Everything works except for arguana corpus; still being debugged.

+ Tweaked scores for ONNX (before, it was scores just copied from pre-encoded version).
+ Completed template docs and links from main README.

Ran four trials on different machines and averaged the scores, to better reflect the variance across multiple trials.

…ge (#2387)

* fix BGE maxlen issue by restricting maxlen to 512
* add maxlen test cases (i.e., long input query) for splade-pp-ed and bge

+ maxlen issue with ONNX fixed in #2387
+ this PR simply copies scores from the non-ONNX versions, will circle back and readjust scores

+ Tweaked scores by averaging over 4 trials
+ Added documentation for Cohere DL19/20

(e.g., when performing runs on pre-built indexes)

… BGE (#2396)

…#2399)

Respond with "OK" if within 0.005; respond with "OK-ish" if within 0.01 (means, take a closer look to verify).

…dings (#2401)

Additionally, rename v3 to v3.0 to be consistent with BGE v1.5.

ref: castorini/anserini-tools#73 castorini/anserini-tools#75

Add reproduction script and html generation script for MSMARCO V1

…#2411)