feat(search-pipeline): compose search indexing as a pipeline instance

[ddeboer]

Slice 3 of #534, stacked on #560: the new @lde/search-pipeline package that makes search indexing an @lde/pipeline Configurable Pipeline instance.

What it does

searchIndexWriter({ schema, writer }) is the one type-changing step (quad → document), shared across engines – which is why it lives here and not inside an engine adapter:

• buffers each dataset’s extracted CONSTRUCT quads until the dataset flushes;
• frames and projects them with @lde/search’s projectGraph (engine-agnostic);
• writes the documents to the engine writer’s run before forwarding the flush, so an In-place stale sweep never races its own documents;
• passes the run lifecycle through unchanged (run context, per-dataset flush outcome, reset for the dump-fallback discard, commit/abort) – the engine writer’s update mode governs, and the pipeline never branches on it.

With this, a search index rebuild is a plain Pipeline with an extraction SparqlConstructReader and writers: searchIndexWriter({ schema, writer: new BlueGreenRebuild(...) }) – reusing Discovery → Selection → change-gate (skip-unchanged) instead of the bespoke full-rebuild indexer.

Scope notes

• Memory is bounded by one dataset’s extraction and released per flush; nothing accumulates across datasets (pinned by a spec). Bounded entity batches within one huge dataset need the two-level iteration (dataset → entity-URI batches) from Search as a Configurable Pipeline instance: transactional Writer, two update modes, per-reference label sources #534 and are explicitly future work (documented in the README).
• 100% coverage baseline; unit-tested against a fake engine writer – the engine writers themselves have their own testcontainer suites (feat(search-typesense)!: BlueGreenRebuild and InPlaceRebuild writers #560).

Releasing

New package: after merge, the first version must be published manually per the npm-bootstrap-package flow (npm view @lde/search-pipeline → 404 today).

Part of #534 (slice 3 of 4; next: per-reference label sources + facet-by-name search).