v0.9.3: bug fixes and general performance improvement

What's Changed

• chore: create f16 and bf16 test dataset from sift and gist by @eddyxu in #1761
• fix: fix query over f16 knn node and sift benchmark by @eddyxu in #1762
• fix: fix shuffler k-list merge and save GPU trained centroids to file by @chebbyChefNEQ in #1763
• fix: pick up metric type from the existing index for the refine step by @eddyxu in #1764
• fix: switch back to stable sampling, and tune memory usage during IVF shuffle by @chebbyChefNEQ in #1765
• fix: handle nan and zeros gracefully in kmean training by @eddyxu in #1766
• fix: fix latent sample size bug in torch dataset by @chebbyChefNEQ in #1772
• chore: handle nans in GPU training by @eddyxu in #1769
• chore(python): prevent segfault using LanceFragment by @changhiskhan in #1768
• chore: skip flaky test by @changhiskhan in #1774
• feat(python): add with_row_id to fragment scanner by @changhiskhan in #1770
• chore: fix f16 distance computation on CPU by @eddyxu in #1776

Full Changelog: v0.9.2...v0.9.3