v1.19.0

🐛 Bug Fixes

• Fixes for resharding abort logic consistency.
• Corrected reading of deferred source points in copy-on-write moves.
• Prevented data loss when a payload index drop races the background segment flush.
• Fixed abort resharding live-lock.
• Buffered multi-dense offsets store to prevent reload corruption.
• Fixed empty `min_should` with non-zero `min_count` matching everything.
• Fixed optimizer data race, deferring source segment destruction until durable.
• Fixed scalar L2 quantized score shift.
• Fixed snapshot WAL clocks data race.
• Avoided allocator abort on huge limit in random search/sample/scroll.
• Fixed Point Group Search query with large limits or group sizes causing panics.
• Fixed reinitialization failing with "removed all voters" on a removed peer.
• Fixed optimizer recreation dropping sibling shard restarts on error.
• Unblocked optimizer after deleting a named vector.
• Fixed cancelled snapshot recovery dirty marker being left behind.
• Fixed `/readyz` of reinitialized peer waiting on a foreign consensus.
• Preserved empty result shape for zero-limit batches.
• Fixed drive letter parsing in Windows snapshot mappings.
• Fixed `/readyz` false positive on a freshly bootstrapped peer.
• Fixed payload index inconsistency after restart: flush-before-build.
• Stamped payload storage version on append-only clone paths.
• Fixed use-after-free in u8-quantized vector reads on owning storages.
• Returned error for missing point IDs in Query API recommend.
• Reset `first_voter` and pruned address book on first-peer `--reinit`.
• Advanced newest clocks for WAL-replay tail on load.
• Counted trailing deleted offset on mmap sparse reload.
• Clamped WAL replay target to the truncated prefix on shard load.
• Fixed resharding, on queries filter shards on all shard selectors.
• Fixed WAL data race, flushing after copy with `Segment::copy_to_path`.
• Cleaned up stale shard transfers when applying consensus.