feat: strengthen ontology-edit guidance to reuse types over inventing

[Evanfeenstra]

What

Follow-up to #1465. Tightens the ontology-editing guidance so the agent reuses existing types instead of inventing new ones out of thin air, and makes inspecting the current ontology a hard prerequisite before any schema write.

Changes

agent.ts — injected ONTOLOGY_EDIT_GUIDANCE:

• Inspection is now a mandatory first step: never call a create/update/delete tool without having inspected the current schema (get_ontology, with graph_search as a fallback to see how a concept is already modeled).
• Added a reuse-first decision procedure (ordered): match existing types by meaning (synonyms, plurals, broader/narrower terms — e.g. Podcast→Episode, Individual→Person, Org→Organization) → extend an existing type (add attribute/edge) → only create as a last resort, inheriting from the closest existing parent and justifying why.
• Frames type-creation as a heavyweight, last-resort action and calls out near-duplicates as the thing to avoid.

toolsJarvis.ts — ontology_create_type / ontology_create_edge descriptions:

• Strengthened the always-present tool text with the same "LAST RESORT / MUST inspect first / match on meaning / prefer extension" framing (defense-in-depth at call-time).

Autonomy note

Deliberately does not tell the agent to ask the user before creating a type — this keeps autonomous use viable. Callers can still prompt for confirmation if they want it. The separate destructive-op confirmation guardrail (delete type/edge, rename attribute) is unchanged.

Testing

• tsc --noEmit clean