-- pg_turbovec 1.29.0 -- Partitioned-scale cookbook + 1-bit foundation.
--
-- This file is a *reference mirror*. The authoritative install script
-- is generated by `cargo pgrx schema`.
--
-- MINOR bump: additive only, no wire-format change (MetaPageData::
-- version stays 7), NO REINDEX. `ALTER EXTENSION pg_turbovec UPDATE TO
-- '1.29.0';` is sufficient and cannot fail on existing indexes.
--
-- Changes:
--
--   Phase S-0 (partitioned scale toward 1T+ vectors, DOCS ONLY) -----
--   pg_turbovec scales past PostgreSQL's ~11B-row single-table
--   ceiling via declarative hash-partitioning + per-partition
--   turbovec indexes; PG's native Merge Append already does correct,
--   LAZY global top-k across the partitions with ZERO new code
--   (proved: 10/10 overlap vs a single-table exact scan; the merge
--   pulls ~k+probed rows, not k*N). See docs/PARTITIONED_SCALE.md for
--   the cookbook (partition sizing, embarrassingly-parallel
--   per-partition build orchestration, insert routing, per-partition
--   VACUUM/REINDEX) + benches/poc/scatter_gather_partitioned_topk.sql.
--   Unlocks 1-10B vectors TODAY with no AM changes. No new SQL surface.
--   Partition PRUNING for very large N (Phase S-1, the partition-level
--   coarse quantizer) is DESIGNED (doc section 6) but NOT shipped here
--   -- it lands in a later release after real-PG validation.
--
--   1-bit (sign binary quantization) FOUNDATION ---------------------
--   `WITH (bit_width = 1)` is now ACCEPTED by the reloption (range
--   1..=4; 1 = sign-BQ, 2/3/4 = TurboQuant; the GUC default stays
--   2..=4 so 1-bit is opt-in). The exact-heap-rerank floor auto-widens
--   for a 1-bit index (reusing hi_dim_rerank/search_k/oversample). The
--   sign-BQ core (mean-centering -- the mandatory footgun fix for
--   non-zero-centered data -- + degenerate-all-same-sign detection +
--   half-of-2-bit storage) is implemented and unit-tested.
--   NOTE: the 1-bit BUILD path is NOT yet implemented -- a
--   `CREATE INDEX ... WITH (bit_width = 1)` ERRORs clearly
--   ("not yet implemented; use bit_width = 2, 3, or 4"), never
--   half-builds or ships a silent all-ones landmine. The end-to-end
--   1-bit encode/scan path (which needs a wire bump to v8) lands in a
--   later release after full validation. The reloption is accepted now
--   for forward compatibility only.