feat: target-ontology-adapter §6 ManifestTargetAdapter

[deanban]

First concrete TargetOntologyAdapter: YAML/JSON manifest parser, descriptor + per-facet completeness defaults, endpoints block schema for GRAPH_EDGE, synthesized vs supplied context cards, and the sema target load --manifest CLI command. No LLM use.

Tracking: #97. Source: openspec/changes/target-ontology-adapter/tasks.md §6.

Tasks

• 6.1 ParsedManifest model
• 6.2 YAML/JSON loaders (extension auto-detect)
• 6.3 ManifestTargetAdapter.__init__
• 6.4 manifest_version=1 validation
• 6.5 describe() with manifest-implied facet defaults
• 6.6 Per-entity override rules (synonyms, inline terms)
• 6.7 Explicit completeness override
• 6.8 discover_entities() sorted output
• 6.9 load_entity
• 6.10 load_obligation
• 6.11 load_vocabulary_bindings
• 6.12 load_context_card with synthesized fallback
• 6.12a Reject card missing card_version
• 6.12b Reject card carrying card_hash
• 6.13 iter_terms (NotImplementedError for EXTERNAL)
• 6.14 endpoints block schema
• 6.14a Reject GRAPH_EDGE missing endpoints
• 6.14b Dangling endpoint refs caught at normalizer time
• 6.14c Reject endpoint targeting TABLE_ROW
• 6.15 Reject endpoints on TABLE_ROW/GRAPH_NODE
• 6.16 Vocabulary declarations (INLINE/EXTERNAL)
• 6.17 @register_target_adapter for manifest
• 6.18 sema target load --manifest <path> CLI
• 6.19 Minimal-manifest test
• 6.20 YAML/JSON equivalence test
• 6.21 Unknown manifest_version rejection
• 6.22 Missing descriptor rejection
• 6.23 Full DTO coverage test
• 6.24 No-LLM spy test
• 6.25 EXTERNAL vocab -> no terms test
• 6.26 Per-facet completeness defaults table
• 6.27 TargetArtifactKind acceptance + endpoint rejection cases
• 6.27a GRAPH_EDGE endpoint round-trip test
• 6.28 GRAPH_EDGE minimum-viable-row default test
• 6.28a Temporal vs non-temporal edge obligation tests
• 6.29 Vocabulary-binding round-trip test
• 6.30 Inline terms upgrade terms to COMPLETE
• 6.31 Synthesized card test (0.0.0-synthesized)
• 6.32 Card validation rejection tests
• 6.32a card_version bump vs drift tests