feat(agents): add PR review closure templates

[stone16]

Summary

Adds a PR review closure architecture spec and two prompt-only curated agent templates: PR Review Orchestrator and PR Review Worker. The shipped MVP lets one Multica issue own a review window while per-PR workers run in separate sessions and report four-state comment outcomes.

Why

STO-136 is about reducing human touchpoints before PRs are ready. The current GitHub integration mirrors PR metadata and check state, but not review comments, so this starts with a reversible template-based workflow that uses runtime GitHub access and one Multica issue summary instead of adding schema prematurely.

Approach

Implemented cp-01 by adding docs/pr-review-closure-architecture.md, covering the orchestrator/worker boundary, runtime flow, data model, observability, security, alternatives, and checkpoint verification commands.

Implemented cp-02 by adding prompt-only templates under server/internal/agenttmpl/templates/. The orchestrator template owns PR discovery, one-PR-per-worker delegation, no-self-review routing, and the final summary contract. The worker template owns checkout-at-head-SHA, four-state review comment resolution, scoped fixes, and verification evidence. Rejected a first-class GitHub review-comment schema for the MVP because the existing product only mirrors PR metadata/check suites; the template path is reversible and faster to validate.

How I Tested

End-to-end test case table:

Case	Coverage	Evidence
Template registry loads embedded templates	New JSON templates are valid and load through the production embed path	server/internal/agenttmpl/loader_test.go:14
Squad leader claim still receives briefing	Existing coordination path remains available for orchestrator use	server/internal/handler/squad_briefing_test.go:279
Squad assignment and mention routing still work	Existing per-PR worker delegation mechanisms are unchanged	server/internal/handler/squad_assign_trigger_test.go:13, server/internal/handler/squad_comment_trigger_test.go:545

Verbatim test output:

$ cd server && go test ./internal/agenttmpl
ok  	github.com/multica-ai/multica/server/internal/agenttmpl	0.379s

$ cd server && go test ./internal/handler -run 'TestClaimTask_LeaderGetsBriefing|TestCreateIssue_AssignedToSquadEnqueuesLeader|TestSquadMentionTriggersLeader'
ok  	github.com/multica-ai/multica/server/internal/handler	0.455s

$ rg -n "^TODO_DECISION:" docs/pr-review-closure-architecture.md server/internal/agenttmpl/templates/pr-review-*.json
<no output; exit 1>

Rollback Plan

Revert this PR. Blast radius is zero data integrity risk: it only adds documentation and static template JSON. Time to rollback is one revert commit.

Out of Scope

• Persisting GitHub review comments in Multica tables.
• Building a dedicated PR review summary UI.
• Posting automated summaries back to GitHub PR comments.
• Changing squad routing, task queue semantics, or GitHub webhook behavior.

[vercel]

@stone16 is attempting to deploy a commit to the IndexLabs Team on Vercel.

A member of the Team first needs to authorize it.