feat(graph): add the document-vector read seam for revision candidates (#1014)

[jasonssdev]

Summary

PR 4 of the decision-revision-detector chain, which is Phase B slice P1 (stacked-to-main).

• Planning docs: the Phase B re-plan. Production candidates read the document vectors that reindex already stores in vectors.db, and revisions never calls an embedder. The subject pass is dropped (owner decision B3). --include-confidential does not authorize an embed call (owner decision B2). The spec and design deltas are updated and the Phase B tasks are written (11 slices).
• P1, the vector read seam:
  • VectorStoreDB.document_vectors(concept_ids) returns StoredDocVector rows (vector + content hash) by joining doc_vectors and vector_meta. It is concrete-store only and deliberately not on the VectorStore Protocol, following the precedent of neighbors.
  • The public embedding_tag(model) in state/reindex.py; _effective_model_tag delegates to it.

Nothing is wired to a verb yet; the seam is inert until P5.

Related issue

Refs #1014

Type of change

• feat — new feature
• docs — documentation only

How was this tested?

• Strict TDD. RED was observed for both behaviors (AttributeError on the missing method and function). 5 mutations were run: 4 were killed. The 5th, removing the empty-input guard, survives because SQLite's IN () matches no rows anyway; the guard is kept as a query-avoidance shortcut.
• ruff check ., ruff format --check . and mypy . are clean; pytest --cov 6823 passed, 97.07% coverage; evals/run_self_tests.py 43 of 43.

Checklist

• My commits follow Conventional Commits.
• I added or updated tests for the change.
• I updated docs where behavior, interfaces, or the knowledge model changed.
• Lint, format, type check, and tests pass locally (ruff, mypy, pytest).
• Output remains OKF-conformant and derived stores stay reconstructible from the bundle + sources.
• The change is consistent with the project's guiding principles (local-first, provenance, freshness, human-in-the-loop).