Align property/attribute naming with the glossary

`CONTEXT.md` defines **Property** as the canonical term for a key-value pair on a Vertex or Edge, and lists _Avoid_: Attribute as "legacy code term being phased out". The code has not followed: `attribute` is still the pervasive spelling across the connector and schema layers.

Current uses of the legacy term include:

- `AttributeFilter` / `NeighborsRequest.attributeFilters`
- `KeywordSearchRequest.searchByAttributes`
- `useSearchableAttributes`, `isAttributeSearchable`, `DisplayConfigAttribute`
- `AttributeConfig`, `VertexTypeConfig.attributes`
- `Vertex.attributes` / `Edge.attributes`

Renaming these one at a time makes things worse rather than better — a half-migrated boundary means a reader has to know both vocabularies and which layer uses which. This task is to do the sweep coherently, or to decide the glossary should adopt "Attribute" instead and update `CONTEXT.md`.

Two things to settle before starting:

- `Vertex.attributes` and `Edge.attributes` are part of the persisted graph session and styling shapes, so a rename there needs backward-compatibility handling for data already in IndexedDB. Check whether the rename should stop at the boundary of persisted types.
- The RDF vocabulary calls these "Datatype Properties", which is where the glossary's preference comes from. Confirm the term reads correctly for property-graph users too before committing to it.

Note the recent `AttributeFilter` type (from #2031) deliberately used the "attribute" spelling to match the surrounding code rather than pre-empt this decision, so it is part of the same sweep.

## Related Issues

- Follows #2031

> [!IMPORTANT]
> Internal only — this issue is maintained by the core team and is not accepting external contributions.

Start with CONTEXT.md and the usages listed in the issue, including AttributeFilter from #2031; map the connector/schema boundary and persisted Vertex/Edge data in IndexedDB before choosing Property or Attribute. Done means the terminology decision is documented and the affected names are swept coherently without breaking existing persisted sessions.

Written by the indexing model from the issue text.