v9.0.0-beta.22

What's Changed

Breaking Changes 🛠

• feat(fts)!: add configurable posting block size by @BubbleCal in #7466

New Features 🎉

• feat(lance-linalg): runtime SIMD dispatch for pre-Haswell x86_64 from-source builds by @tobocop2 in #6630
• feat: data overlay files — model, feature flag, and write/commit path by @wjones127 in #7535
• feat(fts): impact skip data for posting lists by @BubbleCal in #7602
• feat: support list blob scans by @Xuanwo in #7664
• feat(credentials): vend AWS credentials via AssumeRoleWithWebIdentity to avoid role chaining by @rpgreen in #7757
• feat: support multi-segment indices in hamming clustering by @jackye1995 in #7758
• feat: support wider hamming hashes by @jackye1995 in #7767

Bug Fixes 🐛

• fix(index): keep HNSW IVF scratch dir alive during partition writing by @geserdugarov in #6980
• fix(encoding): honor zstd compression scheme and level on large per-value buffers by @claydugo in #7460
• fix(python): preserve PQ num_bits in model training by @ddupg in #7583
• fix(scanner): honor batch_readahead to bound v2 scan decode concurrency by @zhangyue19921010 in #7632
• fix: preserve case in generated field path expressions by @skyshineb in #7698
• fix(encoding): decode empty inline-bitpacked blocks by @Xuanwo in #7752
• fix: train IVF indexes on fragment subsets by @jackye1995 in #7768
• perf: skip fragment-reuse index for compactions with no indexed data by @xuanyu-z in #7774

Documentation 📚

• docs(index): document public API of LogicalScalarIndex by @westonpace in #6804
• docs: specify data overlay files for the table format by @wjones127 in #7381
• docs: clarify Lance format stability by @Xuanwo in #7475

Performance Improvements 🚀

• perf(fts): bulk MAXSCORE search path for top-k disjunctions by @BubbleCal in #7603

Other Changes

• refactor(encoding): clarify mini-block repdef budget by @Xuanwo in #7751

Full Changelog: v9.0.0-beta.21...v9.0.0-beta.22