RAG: chunking + indexing pipeline (embed, upsert, re-index, delete)

[qiuethan]

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

Scope

• Chunk extracted document text (the full-text storage landed in Content acquisition: extend FetchResult + storage for full extracted text (precursor) #83) with a configurable size/overlap strategy.
• Embed each chunk via /embed and upsert into the vector store.
• Re-index on document change; remove chunks when a document is deleted or its source disappears.
• Idempotent: re-running over an unchanged document must not duplicate chunks.

Acceptance

• A document with real extracted text (e.g. from the Google Workspace connector, Google Workspace content connector (Docs/Drive/Sheets/Slides) #84) is chunked, embedded, and persisted end to end.
• Editing the source and re-indexing replaces chunks rather than accumulating them.
• Deleting a document removes its chunks.

Open question

Chunk size / overlap for the doc shapes we actually have — Google Docs, Notion pages, GitHub READMEs. See the epic.