Close frontend review findings

[pandemicsyn]

Summary

• close all five retained P3 frontend-review findings: tabular metric numerals, navigation timer cleanup, operation-scoped PR review status, stable report-finding provenance, and strict Flue agent configuration
• persist deterministic finding IDs on generated draft comments with a backward-compatible SQLite migration and legacy matching fallback
• add focused regression coverage and a patch changeset
• reconcile the frontend review and diff improvement plans while retaining every original finding and marking completed work in place

Why

This is the closing implementation pass for the frontend static review. The remaining issues could cause visual jitter, stale or misleading operation feedback, timer races, incorrect finding identity, and configuration that did not match the exposed agent and session routes.

Impact

The PR review workbench now keeps feedback scoped to the latest operation and tracks generated findings independently of editable display text. Shared live metrics render steadily, navigation pulses clean up correctly, and unsupported Flue agent names fail safely with a visible configuration issue.

Validation

• lint, dependency-layer checks, app and docs type checks
• 551 unit tests, 33 git tests, and 126 integration tests
• web, server, and docs production builds
• npm package validation and packed CLI smoke test
• migration consistency, Prettier, staged diff, and frontend detector checks

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e99f4b79c2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Keep recognizing drafted report-only findings

When finding.sourceId is truthy, this branch stops considering the generated body, but readPrReviewRow() now synthesizes sourceId for older report_only_findings_json rows whose existing draft comments still have source_finding_id null. After upgrading, even an already-generated legacy comment is treated as not drafted (and cross-file manual anchors also miss because of the path guard), so the sidebar re-enables “choose line” and can create duplicate draft comments for the same finding.

Useful? React with 👍 / 👎.