Mention edges readable over the engine seam, both directions (nxf 6j6v.q29p)

[cabcookie]

What

An embedding product can now read mention edges back over its seam, in both directions.

• Engine::mentions(id) -> Result<Vec<String>> returns the outgoing targets. It is the existing read::mentions, put on the seam.
• Engine::mentioned_by(id, &MentionedByQuery) -> Result<NextPage> and Engine::mentioned_by_value(id, &MentionedByQuery) -> Result<Value> return the live items that mention a target, in id order. Closed and archived mentioners count; deleted ones do not. The query takes a NextFilter (type, container = belongs_to OR contributes_to, label) and pages exactly like next_query.
• CLI: nxf mention by <id> [--type T]... [--in ID] [--label L] [--limit N [--paginate [--token T]]].
  • --json gives list --json records (with presentation labels) as a plain array.
  • With --limit it gives {"items", "total"}, and with --paginate also next_token and restarted. This is the same flag-shaped rule as next.

Shared with x0wf (spec derived-ranking-keys section 11)

read::mentioners_of(store, targets: &[&str]) -> Result<BTreeMap<String, Vec<ItemRow>>> is the reverse lookup:

• For each target it gives the live mentioners, id-sorted, in one edge query per 500 targets (Store::mentioned_by_bulk, new in core) plus one item read.
• mentioned_by narrows its result with a NextFilter.
• count = "mentioned_by" should count the rows per candidate that pass its type/status/label parts. That costs O(E + N) over a candidate set (spec section 5).

Paging reuse, not a copy

The paging half of next_page_at is factored into the private Paging (validation), Watermark and page_through (cut, resume, cache).

• next_page and mentioned_by_page both call it, so they share the next_page_cache table, the 1h expiry, the 8-per-query and 64-total caps, the invalidation rules and the restart-on-stale-token behaviour.
• The mentioned-by snapshot key names the read and the target ({"read":"mentioned_by","target":..,"filter":..}), so a token never crosses between next and mentioned_by or between targets. next's key format is unchanged, so existing tokens keep working.
• The table keeps its name; no schema change.

Tests

• crates/facade/tests/mentioned_by.rs (new, 12 tests at the Engine seam): outgoing read, live/closed/deleted/removed set, not_found, type and container filters, limit, paginate walk, restart on a removed mention, outside op that does not invalidate, cross-target/filter/read tokens restart, validation, value envelope, bulk mentioners_of.
• read.rs unit test: a mentioned-by token expires on the injected clock.
• store.rs unit test: mentioned_by_bulk.
• crates/cli/tests/mention.rs: 4 black-box tests (records + filters, envelope + paging + stale restart, human view, errors). $HOME is pinned via cargo_bin.
• Goldens under both plugins; guide EN/DE quotes them.
• Verb-seam gate: Alias mention by -> mentioned_by.

Run locally: those test files, next_query (18, unchanged), the facade lib paging tests, verb_seam, golden, guide_examples, cargo xtask docs check / docs coverage, cargo xtask changelog check. No clippy and no full suite were run locally; CI is the gate.

Fragment: changes/q29p-mention-reads.md, facade: changed (additive).

🤖 Generated with Claude Code

https://claude.ai/code/session_016q5uMUkZJQW4fckJoA3FME

[cabcookie]

Contained in #49, which folds #46, #47 and #48 plus the review mitigation and was merged in one squash.