-- pg_turbovec v1.26.0 -- Phase G-2d(a): partitioned/merge parallel
-- build for the graph index kind.
--
-- This file is a *reference mirror*. The authoritative install script
-- is generated by `cargo pgrx schema`.
--
-- NO wire-format change (stays VERSION 6, byte-identical on-disk CSR).
-- Existing v4/v5/v6 indexes all decode unchanged. NO REINDEX.
--
-- New SQL surface (additive, minor bump): one new GUC,
-- `turbovec.graph_build_partitions` (int; default 0 = auto). No new
-- operators, types, functions, or opclasses.
--
-- What it does (build-time only): the single-pass Vamana build for
-- WITH (graph = true) indexes is serial by necessity and did not
-- complete at 5M rows. This adds
-- a structurally-parallel build: partition the corpus into P shards
-- (contiguous ranges of the deterministic shuffled insertion order),
-- build each shard's sub-graph in parallel across the bounded rayon
-- pool, then stitch via a parallel cross-shard refinement pass
-- (greedy-search the merged graph from a global medoid entry +
-- RobustPrune per node) and a deterministic parallel reverse-edge
-- pass. It emits the IDENTICAL on-disk CSR shape a single-pass build
-- does -- no persisted bridge edges, no wire change.
--
-- Measured: recall parity (partitioned matches or BEATS single-pass:
-- 0.958 -> 0.996 R@10 in a findable regime), ~8x parallel build
-- speedup (P=16, 200k rows, 8-core box), bit-identical determinism
-- across (corpus, seed, P) and across rayon pool sizes.
--
-- turbovec.graph_build_partitions: auto (default) derives P from
-- corpus size + build-pool budget (single-pass below a threshold);
-- 0/1 forces single-pass; N forces N shards.
--
-- Migration: `ALTER EXTENSION pg_turbovec UPDATE TO '1.26.0';`. No
-- REINDEX -- existing graph indexes are unaffected; the parallel build
-- only changes how NEW graph indexes are constructed, to the identical
-- on-disk shape.