Finish the unified PR comment (close out PR #1007)

[JSONbored]

Part of #983.

Context

The unified PR comment is natively re-ported into gittensory (src/review/unified-comment.ts + src/review/unified-comment-bridge.ts, zero reviewbot imports) and wired into the OPEN-PR path only, behind env.UNIFIED_REVIEW_COMMENT (default "false"). The CLOSED/skipped-PR path is still unconditional legacy: maybePublishPrPublicSurface (processors.ts:1339) posts buildClosedPrPanelUpdate(repoFullName, pullNumber) (processors.ts:1284) — a hardcoded > [!NOTE] Gittensory Gate skipped table — with NO flag check, so with the flag ON an open PR's unified comment is overwritten by the legacy panel on close (same marker gittensory-pr-panel:v1). There is also NO processGitHubWebhook → maybePublishPrPublicSurface → comment integration test. This closes out PR #1007.

Deliverables

• Route the closed/skipped-PR panel (buildClosedPrPanelUpdate, processors.ts:1284, posted at :1339 with createIfMissing:false) through the unified renderer behind isUnifiedReviewCommentEnabled; the bridge gateConclusionToVerdict already maps skipped → comment, so add a closed-state branch in unified-comment-bridge.ts.
• Single-source PR_PANEL_COMMENT_MARKER — it is duplicated as a literal in unified-comment-bridge.ts:34 vs the canonical comments.ts:4; import the one from comments.ts and confirm the legacy marker aliases (gittensory-pr-intelligence, gittensory-agent-command, comments.ts:7-8) still upsert in place (comments.ts:68 markerAliases).
• Add the FIRST processGitHubWebhook → maybePublishPrPublicSurface → comment integration test (none exists; use TestD1Database, test/helpers/d1.ts) asserting: flag-OFF = legacy body, flag-ON = unified body, exactly ONE marker present, identical Gittensory Gate conclusion both ways.
• Verify the AI consensusDefect is not shown twice (once as a gate-blocker row, once as a Code-review blocker) when the flag is ON — runAiReviewForAdvisory mutates advisory.findings with ai_consensus_defect BEFORE the gate, and the unified comment surfaces it again as a reviewer blocker.
• Confirm both comment branches are public-safe: neither the unified nor the legacy branch calls sanitizePublicComment at processors.ts:1647 today (relies on per-input construction safety — AI notes via toPublicSafe, gate blockers via sanitizeForCheckRun, panel rows via sanitizePanelText). Prove construction-safety for every field fed into buildUnifiedCommentBody, OR add a sanitize pass.
• Flip UNIFIED_REVIEW_COMMENT to SHADOW on one repo.

Acceptance criteria

• Open and closed PRs render through the SAME unified renderer when the flag is ON; the legacy buildClosedPrPanelUpdate no longer surfaces under the flag.
• Exactly one gittensory-pr-panel:v1 marker per PR; the comment upserts in place (no duplicate posts).
• The new integration test passes for both flag states and asserts an identical Gittensory Gate conclusion.
• No private framing (reward/score/wallet/rubric/path/threshold) can leak into either comment branch.

Notes

• Milestone: M1
• High-rebase-collision risk: codex churns processors.ts (maybePublishPrPublicSurface), comments.ts, and signals/engine.ts across other gittensory worktrees. Keep convergence edits minimal + flag-gated and rebase onto main before in-flight codex PRs touching those files settle.
• Coordinate logging with the Sentry seam (Sentry: privacy scrubber (beforeSend redaction) — non-negotiable #1000/Sentry: review-pipeline tracing + per-installation tagging #1001) so private config never lands in error reports.