RAG: pgvector store + embeddings schema (documentation-system)

[qiuethan]

Sub-issue of the RAG retrieval epic (#125). Stage 2 of 3: index.

Scope

• Enable the pgvector extension on the documentation-system database via an Alembic migration.
• Embeddings table keyed to document + chunk, storing the vector, chunk ordinal, and source offsets.
• Record the embedding model and dimensionality per row so a model change is detectable rather than silently corrupting similarity.
• Index appropriate for the expected corpus size.

Acceptance

• Migration applies cleanly on a fresh database and in CI (documentation-system-test already runs migrations against Postgres).
• A vector can be written and read back with its document/chunk linkage intact.