Release v0.16.0 · lance-format/lance

What's Changed

feat: standalone vector transform stage by @westonpace in #2566
feat: coalesce scheduling of reads to speed up random access by @raunaks13 in #2636
feat: add version tags by @dsgibbons in #2482
feat: object store registry for custom object store providers by @maxburke in #2513
feat: merge_insert update subcolumns by @wjones127 in #2639
feat: implement disk-based inverted index by @BubbleCal in #2643
feat: allow round-tripping of dictionary data through the v2 format by @westonpace in #2656
feat: support transforming selected fragments in vector transform stage for ivf_pq index by @raunaks13 in #2657
feat: expand tags api by @dsgibbons in #2679
feat: add standalone shuffle for transformed ivf-pq vectors file by @raunaks13 in #2670
feat: support bitpacking for signed types by @albertlockett in #2662
feat: support loading huggingface image dataset and convert image to PIL by @eddyxu in #2684
feat: return BM25 scores for FTS by @BubbleCal in #2687
feat: add data file format / version information to manifest by @westonpace in #2673
feat: add support for the null data type to v2 by @westonpace in #2685
feat: add backpressure to v2 I/O scheduler by @westonpace in #2683

fix: correctly encode a list type when all items are empty by @westonpace in #2653
fix: improve error message when can't train PQ on too small dataset by @albertlockett in #2644
fix: slight cleanups to path handling so that the indices builder tool properly supports Windows by @westonpace in #2689

docs: schema evolution by @wjones127 in #1911
docs: reorg the scalar index python docstring to make the index type clear by @eddyxu in #2678

perf: add v2 fragment file metadata to the FileMetadataCache by @jiachengdb in #2647
perf: add random take benchmark by @chebbyChefNEQ in #2654
perf: benchmark lance vs parquet read time, write time, and compressed size by @raunaks13 in #2383

Full Changelog: v0.15.0...v0.16.0