v2 tracking: plan §9 deferred features

[deOliveira-R]

Umbrella issue for plan §9 items intentionally deferred from v1. Each has its own thread once work starts.

Checklist

• CSL citation styles beyond BibTeX. Wire citeproc-py so the detail page can render APA / IEEE / Nature / Chicago / etc. The MCP `get_citation` tool should also accept `style` beyond `bibtex`.
• Annotation import/export (Zotero JSON). Round-trip highlights and notes with Zotero's annotation schema. Useful if Rodrigo ever co-reviews with non-grimoire users.
• Tag hierarchies. Extend `tags` table with `parent_id` (same pattern as `collections`). Schema change — needs explicit approval (plan §10 rule 5).
• Author disambiguation beyond normalized-key + ORCID. Add an "also-known-as" merge table and a human-in-the-loop review queue for ambiguous `J. Smith` / `John Smith` / `Jane Smith` collisions.
• Browser bookmarklet for one-click URL ingest. JavaScript that calls `POST /submit-url` (see v2: upload + URL-submit web forms (deferred from v1.0) #12) with the current page URL.
• OCR batch as background job. `grimoire ocr --since YYYY-MM-DD` that walks items whose extracted text is empty / degenerate and runs Tesseract. Slow (30s–2min per page); must not block the ingest path.
• Multi-device sync. Explicitly out-of-scope for v1 per plan §2 non-goals; list here only so it doesn't get re-raised silently.
• Custom schemas for INL reports / standards. Domain-specific item type + extra fields. Only if Rodrigo's library actually has enough of them to justify.

See plan §9 in `codex_implementation_plan.md` for the original rationale.