Migrate rpt's Claude workflows to the Morrison-Lab/gha reusable workflows

[d-morrison]

Migrate rpt's Claude workflows to the Morrison-Lab/gha reusable workflows

rpt's .github/workflows/claude-code-review.yml and claude.yml are bespoke, standalone anthropics/claude-code-action@v1 workflows. Other lab repos (ai-config, serodynamics, ...) instead consume the reusable workflows in Morrison-Lab/gha, which inherit all the hardening (stub-review retry, verdict guard, dedup of prior-review context, older-comment collapse, cost comment, and fixes like gha#400) automatically as @v2 slides.

Migrating rpt is a DRW win: one less bespoke workflow to maintain, and rpt stops missing fixes that land upstream. But per gha's own "diff feature-by-feature, don't infer parity" rule, it is not a drop-in -- two gaps need decisions first:

Gap 1: tag mode vs agent mode

rpt's review workflow sets track_progress: true on pull_request (tag mode), which posts a live tracking comment and enables the inline-comment tool. The gha reusable workflow defaults to agent mode (track-progress: false, summary-only reviews) and deliberately discourages tag mode, because tag mode hard-grants git-write tools regardless of --disallowedTools (caused unauthorized commits on serodynamics#175). Options:

• Accept agent-mode summary-only reviews for rpt, or
• Wait for / drive anthropics/claude-code-action#1415 (a read_only input) so tag mode is safe, then set track-progress: true.

Gap 2: reviewer toggling

rpt's workflow removes the d-morrison review request while Claude runs and re-adds it afterward (the "restore reviewers after work" pattern). The gha reusable review workflow has no equivalent. Options:

• Add reviewer-toggling as a feature of the reusable workflow (an input), or
• Keep it as separate caller-side jobs around the uses: call.

Scope

• claude-code-review.yml -> Morrison-Lab/gha/.github/workflows/claude-code-review.yml@v2 (passing rpt's R-package prompt as prompt-addendum, allowed-bots: "github-actions[bot],claude").
• claude.yml -> the gha claude.yml reusable agent workflow (separate migration; same reviewer-toggling gap applies).

Filed as a deliberate follow-up; surfaced while using rpt as the live test bed for gha#400 (the duplicate-review-comment fix).

[d-morrison]

Working on this --- paws off until I'm done.

Two research findings that shape the migration:

• Gap 2 (reviewer toggling) is already resolved upstream. The reusable claude-code-review.yml@v2 now stashes/clears all requested reviewers (human and AI, e.g. Copilot) before Claude runs and re-requests them when it finishes --- a superset of this repo's hardcoded d-morrison-only toggle. So no caller-side reviewer jobs are needed; it comes for free.
• Gap 1 (tag vs agent mode): upstream anthropics/claude-code-action#1415 (the read_only input) is still an open, unmerged feature request, and agent mode can't post inline comments (anthropics/claude-code-action#635 leaves the inline-comment tool uninitialized). Decision (confirmed with the maintainer): migrate now in agent mode, accepting summary-only reviews for now --- matching rme's migration --- and flip track-progress: true later once either upstream fix ships.

Scope of this PR: claude-code-review.yml only. Per this issue's own note, the claude.yml agent-workflow migration is a separate follow-up (rme kept its bespoke claude.yml, so there's precedent either way).