feat(unic-archon-dlc): /pr-review finds as much as pr-review-toolkit, without vendoring it #629

Problem

/pr-review finds less than pr-review-toolkit does. ADR 0026-pr-review-generic-archon-harvest first gave the Box seven aspects, taken from the toolkit through unic-pr-review. § 8 of that ADR then folded them into the code-review Method's two axes (#281), so that upstream maintains the review criteria and this repository does not. The maintainer wants the Box to find as much as the toolkit does, without vendoring it, so that it keeps upstream's updates. This reverses ADR 0026 § 8.

#547's resolution, § Broadening /pr-review, not decided here holds the maintainer's three decisions, and the table that compares spawning the toolkit's agents by subagent_type with invoking /pr-review-toolkit:review-pr.

Order

This issue waits for three things:

• #628, the spike, which measures whether an Archon node can reach a plugin agent, command or skill at all.
• #501, the reviewer's variance baseline on 0.28.0, taken before any fix lands.
• #502, run 4, whose predictions are sealed against the current reviewer.

It also touches #482, #486 and #487, which read the review node's output. Its grilling decides the order against each of them.

Acceptance criteria

To be written at grilling, from the spike's result. The grilling decides, and writes into this body:

• Which of the two ways the Box uses, or why neither.
• How the Box keeps its finding contract (confidence, hash, anchor, reconcile) and the Intent Brief.
• Whether ADR 0026 § 8 is superseded by a new ADR.
• Which diagram under apps/claude-code/unic-archon-dlc/docs/architecture/ the change makes stale, with its JSON updated and both PNGs exported again.

The v1 /pr-review sits beside a Consumer's existing reviewer and does not replace it (#547, Q6).