feat: make PR review automation agent-agnostic

[AnrokX]

Summary

• Added agent-agnostic PR review automation support in both automation service and queue/manual launcher flows.
• Wired reviewer spawn/feedback to preserve selected agent session identity (Claude or Codex) and added session tracking for precise feedback routing.
• Added configurable Codex/Claude reviewer options to defaults and documentation.
• Updated Queue panel actions to spawn review/fix/overnight/recheck agents using the configured task launch agent preference.

Files

• client/app.js
• server/prReviewAutomationService.js
• user-settings.default.json
• CODEBASE_DOCUMENTATION.md

Notes

• Reviewer selection now includes both -claude and -codex sessions and labels messages by actual agent.
• PR automation config now supports:
  • reviewerAgent, reviewerMode, reviewerProvider
  • reviewerSkipPermissions (Claude)
  • reviewerCodexModel, reviewerCodexReasoning, reviewerCodexVerbosity, reviewerCodexFlags (Codex)

[web3dev1337]

Assessed in an open-PR sweep: the reviewer-spawn architecture here (bespoke per-agent reviewerCodex*/reviewerClaude* keys) is the pattern the codebase has since moved away from — the agentManager registry + agentSpawnHelper design (PR #1022 lineage) covers agent-agnostic spawning generically, and merging this would revert main's applyWorktreeSessionsAddedPayload refactor in client/app.js (structural conflicts there). HOWEVER: the review-routing/paste-back feature (reviewSourceSessionId linking, latestReviewSummary cache, open/paste-review buttons) is genuinely novel — nothing on main or #1022 duplicates it. Recommendation: CLOSE this branch, and reimplement the paste-back feature fresh on top of the agentSpawnHelper architecture.

[web3dev1337]

Closing per the sweep assessment — the reviewer-spawn architecture is superseded by the agentSpawnHelper/agentManager registry design. NOTE for future work: the review-routing/paste-back feature in this PR (reviewSourceSessionId linking, latestReviewSummary cache, open/paste-review-to-source-terminal buttons) is genuinely novel and unduplicated — worth re-implementing fresh on the current architecture.