Advisory PR-review workflow: trim synchronize trigger, pin model

[smartwatermelon]

Problem

The advisory (non-blocking) claude-code-review.yml pattern — companion to
claude-blocking-review.yml — fires on pull_request: [opened, synchronize, ...], i.e.
every push, not just PR open, and (per current knowledge) has no model: input pinned,
unlike the blocking workflow which was deliberately pinned to
claude-haiku-4-5-20251001 in claude-config commit #27.

Local run-review.sh and the blocking CI review already cover every commit/PR-open, so an
advisory re-review on every incremental push mostly restates prior findings — the highest-
frequency, lowest-marginal-value spend in the pipeline per Anthropic's
model/effort guidance.

Not re-litigating the decision to keep the blocking review (see claude-config
CI-review-strategy memory) — this is scoped to the advisory workflow only.

Proposed fix (pick one or both)

• Drop synchronize from on.pull_request.types for the advisory workflow — review once
  per PR lifecycle event that changes reviewer scope, not every incremental push.
• Pin model: claude-haiku-4-5-20251001, matching the blocking workflow.

Scope note

Need to confirm exactly where this workflow file lives (reusable workflow here in
github-workflows, vs. per-repo .github/workflows/claude-code-review.yml in app repos) —
it is not present in smartwatermelon/dev-env's workflows.

Reference

Full analysis: dev-env repo, docs/plans/2026-07-31-cost-perf-followups.md, item 2.

Claude-Session: https://claude.ai/code/session_017PUAfxBAZPUSyt3nYzwqcq