fix: surface publish submission status

[jcfischer]

Closes #196.

Plan artifact:

• Adds Plans/2026-06-03-arc-issues-196-183-184-sequential-plan.md for the requested publish: surface MetaFactory submission status instead of unconditional Published #196 -> arc install/upgrade re-appends provides.hooks non-idempotently → duplicate PreToolUse registrations #183 -> arc upgrade: extracted-tarball install + version-check both broken (soma#127 reproducer) #184 queue.

Changes:

• Preserve submission_id and submission.status from MetaFactory version registration responses.
• Preserve version.id for the current server response shape when version_id is absent.
• Show pending review publishes as uploaded and queued for review instead of Published.
• Treat rejected publish submissions as failures and include the review comment.

Validation:

• rtk bun test test/unit/publish.test.ts
• rtk bun test test/commands/publish.test.ts
• rtk bunx tsc --noEmit
• rtk bun run lint
• rtk bun test (971 pass, 3 skip, 0 fail)

[jcfischer]

CodeReview — StandardReview (CodeQuality lens)

Auto-detected lenses: CodeQuality (always) — no security surface, auth, or architecture changes detected.

Assessment

Clean, well-scoped fix. Three changes working in concert:

1. RegisterResult type (src/types.ts) — extended with submissionId + submission object to capture server review state. reviewComment correctly typed as string | null matching the server shape.

2. registerVersion() (src/lib/publish.ts) — response parsing now captures submission_id, submission.{id,status,review_comment}. Defensive ?? fallback for version_id vs version.id handles API shape evolution gracefully.

3. publish() + formatPublish() (src/commands/publish.ts) — rejected submissions return success: false with clear error. Pending review intercepts before the "Published" message. Other states (approved, no-submission) fall through to existing behavior. No regression path.

Test coverage: 3 new tests (unit: response parsing, command: pending_review format, command: rejected with review comment). All assert both the result object and formatted output. Good boundary coverage.

No findings. blockers=0 majors=0 nits=0

Recommend: approve.