Typed entity collections, federated search, facet-by-name

[ddeboer]

Typed entity collections, federated search, and facet-by-name for the search-as-pipeline design (epic: #534). Depends on the composition package.

Scope

• Typed entity collections (Multi-source Composition: same kind + many sources -> one collection; different kinds -> separate collections; kind and source orthogonal; identity is URI equality).
• Cross-dataset identity: per-dataset docs keyed by (source, URI) (keeps In-place per-source sweep clean); standalone entity search dedups at query time via group_by on the URI. No index-time merge; the representative (or a field merge) is chosen in the resolver.
• Cross-type federated search (multi_search across schemas).
• Facet-by-name (v1): typeahead the reference target collection by label -> user selects -> filter the main query by the bounded selected IRIs; contextual counts on debounced keystroke (query-time, respecting active filters + skip-own-filter; not precomputed).

Acceptance

• Standalone entity search returns one doc per shared URI (group_by).
• Reference-facet typeahead returns matches with contextual counts; selecting filters the main query.

Design and rationale: #534.