v0.13.0

What's Changed

Breaking Changes 🛠

• fix: dot distance so kmeans can converge by @eddyxu in #2509

New Features 🎉

• feat: convert binary logical encoding/decoding to physical array encoding/page decoding by @raunaks13 in #2426
• feat: support stable row ids in Dataset::take_rows() by @wjones127 in #2447
• feat: enable aarch64-pc-windows builds by @NickDarvey in #2512
• feat: expose DatasetPreFilter, PreFilter, and FilterLoader to the public API by @westonpace in #2522

Bug Fixes 🐛

• fix: low recall if many partitions with only 1 row by @BubbleCal in #2487
• fix: allow row id only in v2 by @westonpace in #2481

Documentation 📚

• docs: flickr8k_dataset_creation_example by @tanaymeh in #2489
• docs(python): note multiprocessing incompatibility by @wjones127 in #2506

Performance Improvements 🚀

• perf: don't write precomputed partitions to trace file by @westonpace in #2493
• perf: use faster kmean find partition routing for pq assignment by @eddyxu in #2515

Other Changes

• refactor: combined capacity updates and decoding for all physical decoders by @raunaks13 in #2508

New Contributors

• @NickDarvey made their first contribution in #2512

Full Changelog: v0.12.3...v0.13.0