[ClaudiaFang]

Summary

First implementation slice for #32 (part of #31): selectively port evidence-grounded Decision Attribution from b08x into Firstsun's analysis pipeline, without replacing Firstsun schema or UI.

• Analysis prompt and JSON Schema now include decided_by (user | agent | collaborative | unknown) plus optional intent and branch_point.
• Existing User#N / Assistant#N evidence remains; prompt now specifies role-specific zero-based ordinals.
• Persist provenance in insights.metadata with conservative unknown fallback for absent/invalid actor values, not collaborative.
• Expose attribution + decision context in shared Decision UI.
• Bump extraction version to 3.1.0 so the new analytical output is identifiable.
• Add unit tests for agency persistence, legacy fallback and prompt instructions.

No SQLite migration in this slice. Firstsun V1-V14, Multi-home/Personality/Kilo, Docker and Node24/pnpm12 configuration are untouched.

Remaining for #32 (this PR is DRAFT)

• Backend TurnRef -> message ID resolver rejects malformed or out-of-range references; literal quote grounding still requires manual/fixture verification.
• Click-through session evidence uses backend role-ordinal resolution, fetches a bounded page and highlights the message.
• End-to-end fixtures for multi-home, tool-result rows, compactions, duplicate timestamps and actor classification quality.
• CI/build/test passed: https://github.com/firstsun-dev/code-insights/actions/runs/40623946029
• Final technical/code review; no merge or production deploy until approval.

Source

• https://github.com/b08x/code-insights/blob/development/cli/src/analysis/prompts.ts
• https://github.com/b08x/code-insights/blob/development/cli/src/analysis/analysis-db.ts

Refs #31; Partially implements #32.