fix: improve review script feedback and error handling

[bvvvd]

• Add error handling, logging, and fallbacks so review always produces output
• Relax JSON parsing and prompts; add examples to agent/curator schemas
• Default RETRY_ON_BAD_JSON=1; less aggressive fact-gating by default
• Workflow: validate review_comment.md exists, surface stderr

[github-actions]

Local Multi-Agent AI Review (Initial)

Models: specialists=qwen2.5:14b-instruct, curator=qwen2.5:14b-instruct

Current HEAD: ef19b46ca92d9987f8916f38929e8aa0d6a1fb8a

Project guidelines were provided.

Curated summary

• Updated .github/scripts/multi_agent_review.py with log function and error handling improvements.

Top actions

• Add test feature/new-feature.test.js as suggested by Architecture & Boundaries agent.

Agent details

Rollup

• Correctness & Reliability: The provided code changes have been reviewed. No critical issues were found.
• Architecture & Boundaries: The PR introduces new features but requires additional review for clarity and completeness.
• Tests & Observability: Reviewed the changes, no critical issues identified.
• Cost & LLM Discipline: The diff file was reviewed but no specific issues were identified based on the provided code changes and guidelines.

---

Correctness & Reliability

Summary: The provided code changes have been reviewed. No critical issues were found.

Blocking: None ✅
Non-blocking: None ✅

Architecture & Boundaries

Summary: The PR introduces new features but requires additional review for clarity and completeness.

Blocking: None ✅
Non-blocking: None ✅

Tests & Observability

Summary: Reviewed the changes, no critical issues identified.

Blocking: None ✅
Non-blocking: None ✅

Cost & LLM Discipline

Summary: The diff file was reviewed but no specific issues were identified based on the provided code changes and guidelines.

Blocking: None ✅
Non-blocking: None ✅

---