feat: Respect author replies to dismiss reasonable justifications

[joaopereirajp]

Summary

This PR introduces the ability for the bot to read and understand replies to its own review comments. If an author replies to a finding with a reasonable justification explaining why the change is not applicable, the bot will dismiss that finding and stop re-requesting it in subsequent review rounds.

This prevents review loops where the bot repeatedly requests changes for something the author has already explained is not needed.

Key Changes

• Stable Finding IDs: Each ReviewFinding now has a stable id generated from the file, line, and normalized description. This ID is used to track findings across review rounds.
• Invisible Comment Markers: The bot now embeds an invisible HTML comment (<!-- livvie:finding-id:abc123 -->) in every inline review comment. This marker is used to correlate replies back to the original finding.
• New Module review-replies.ts: This module handles the new logic:
  • fetchAuthorReplies: Fetches review comments and identifies replies from human authors to the bot's comments.
  • evaluateAuthorReplies: Uses the configured LLM to evaluate the reasonableness of each author reply.
  • filterDismissedFindings: Removes findings from the review that have been justified by the author.
• Pipeline Integration: A new optional stage ("4.5. Reply Evaluation") has been added to the pipeline, which runs after consolidation and before posting the review.
• New respect-author-replies Input: A new opt-in boolean input respect-author-replies has been added to action.yml to enable this feature (defaults to false).
• Documentation: The README.md has been updated with documentation for the new feature, including a new section explaining how it works.

How to Use

To enable this feature, add the following to your workflow file:

- uses: 4itworks/livvie_code_review@v1
  with:
    # ... other inputs
    respect-author-replies: "true"