feat(generators): add --default-language flag for language-tagged literals

[jdsika]

Summary

Adds a --default-language CLI option to both gen-owl and gen-shacl that emits BCP 47 language-tagged string literals (e.g. "Person"@en) for human-readable annotations.

This enables ontology producers to comply with RDF 1.1 §3.3 (language-tagged strings as rdf:langString) and OWL 2 §6.3 (annotation property values) without manual post-processing.

Problem

LinkML generators currently emit all string literals as plain xsd:string values, even for human-readable annotations like rdfs:label, rdfs:comment, sh:name, and sh:description. This prevents downstream consumers from:

• Filtering labels by language in SPARQL (FILTER(lang(?label) = "en"))
• Supporting multilingual ontologies
• Complying with W3C best practices for language-tagged metadata

The LinkML metamodel already has an in_language metaslot, but no generator uses it.

Changes

gen-owl (owlgen.py)

• New default_language field on OwlSchemaGenerator
• _LANGUAGE_TAGGABLE_RANGES frozenset (string, ncname) guards tagging — technical types (URI, integer, boolean, datetime) are never tagged
• _resolve_language() resolves element-level in_language → generator-level default_language → None
• _literal() helper creates properly tagged Literal objects
• add_metadata() tags string-range and fallback-range annotation literals
• add_enum() PV labels respect language tags (constraint values in owl:oneOf are correctly NOT tagged)
• New --default-language Click CLI option

gen-shacl (shaclgen.py)

• New default_language field with __post_init__ whitespace normalisation
• NodeShape rdfs:label / rdfs:comment get language tags
• PropertyShape sh:name / sh:description get language tags via prop_pv_text()
• _add_annotations() tags string annotation values
• Numeric literals (sh:order, sh:minCount, etc.) are never tagged
• New --default-language Click CLI option

Tests

• 7 new OWL tests: tagged labels, backward-compat plain literals, URI ranges, in_language override, annotations, empty string, whitespace-only
• 7 new SHACL tests: NodeShape, PropertyShape, plain literals, numeric guards, annotations, empty string, whitespace-only

Backward compatibility

• Default is None (no language tags) — existing behaviour is completely unchanged
• Empty strings and whitespace-only values are normalised to None

Standards compliance

Standard	Requirement	Status
RDF 1.1 §3.3	rdf:langString vs xsd:string distinction	✅
OWL 2 §6.3	Annotation properties accept rdf:langString	✅
SHACL §2.3.2.1	sh:name / sh:description range includes rdf:langString	✅
BCP 47 (RFC 5646)	Language tag format	✅ (no pre-validation, consistent with rdflib)

Testing

• 113 tests pass (91 owlgen + 22 shaclgen), 4 skipped
• 3 rounds of adversarial review completed with 0 open bugs