feat: reflect_scars tool + default execution_successful at close

[nTEG-dev]

Summary

• Option A: session_close now defaults execution_successful based on confirm_scars decision — APPLYING scars default to true, unmentioned scars default to false. Fixes the 80% null rate in scar_usage records.
• Option B: New reflect_scars tool (gitmem-rf / gm-reflect) for end-of-session OBEYED/REFUTED evidence with minimum evidence thresholds. Reflections are bridged into scar_usage at close with priority over Option A defaults.
• Closes the scar enforcement gap where agents confirm scars but never report outcomes, causing misleading 33% success rate in analytics.

Changes

File	What
src/tools/reflect-scars.ts	New tool — validates reflections, persists to session state + file
src/tools/session-close.ts	Bridges reflections into scar_usage; defaults execution_successful
src/tools/definitions.ts	Tool definition + gitmem-rf / gm-reflect aliases
src/server.ts	Routing + help entry
src/services/session-state.ts	addReflections() / getReflections()
src/services/metrics.ts	reflect_scars in ToolName + performance target
src/types/index.ts	ScarReflection, ReflectScarsParams, ReflectScarsResult types

Test plan

• Build passes (tsc clean)
• All 865 tests pass
• Manual E2E: session_start → recall → confirm_scars → work → reflect_scars → session_close — verify scar_usage records have execution_successful populated
• Verify Option A defaults work when reflect_scars is NOT called
• Verify Option B reflections override Option A defaults when both exist

Refs: OD-772

🤖 Generated with Claude Code