Feat: Add feedback loop mechanism

[HungKNguyen]

Currently the context being passed to the Claude Code review contains only the PR details as well as the PR diff. In situation where the bot made a false positive finding and posted it to the PR, there is no mechanism for a human to appeal the finding. As a result the bot will continuously flag the same issue on subsequent reviews, and or keep requesting changes for the PR (see example).

This PR address that problem by adding a review_context to the review prompt. This review contexts contains:

• Previous findings the bot made on this PR
• Reply comments for each of findings, of which this will be the main mechanism in which a human can give a feedback
• In coporate extra information like the thumb up/down ratio and reactions so that the bot can factor that in in future reviews
• Prompt instruction for how the AI should use this information (take human feedback as an input and challenge the human if they gave invalid justification)

Ps. Wasn't sure that is the original intention of adding a thumb up and a thumb down in the add finding scripts (scripts/comment-pr-findings.js:57). My best guess is that the bot is trying to seed reaction UI for the user

[github-actions]

Found 1 correctness issue. Consider addressing the suggestions in the comments.