[Demirrr]

Refs #571 (leaving it open for the other suggestions: materialisation / entailment-aware SPARQL).

instances(ce, direct=False) only expanded rdfs:subClassOf* for the root variable (?x a), so fillers such as ?s_1 a :Grandparent in Grandfather ⊓ ∃hasChild.Grandparent missed individuals typed only by a subclass.

• New with_subclass_closure helper expands every ?var a <Class> triple, each with its own variable (keeps the Issue with indirect instance retrieval on TripleStore #575 scoping fix); skips owl:NamedIndividual.
• instances, instances_count and is_subset_of_individuals all use it; the latter two previously shared one ?some_cls across scopes.
• Tests: filler expansion, NamedIndividual skip, count/subset scoping, and a semantic check running the rewritten query with rdflib on a tiny graph. They fail on the old behaviour.

Cost: each filler adds a subClassOf* path join, so indirect queries on deeply nested expressions can be slower on large stores. direct=True is unchanged.

Tested locally with the mocked triplestore tests only (no live Fuseki); the triplestore CI job is the first real-Fuseki run.

🤖 Generated with Claude Code