review: test same-model/different-thinking inspectors link to distinct sessions

normalizeReviewModelID strips the thinking-level suffix, so two inspectors on the same agent with claude-sonnet:high / :low normalize to the same model id. Add a manifest test with both sessions starting in the same second to pin that used-session tracking in buildLocalReviewManifestFromSummary still links each worker to a distinct session (without it, both would grab the same most-recent match).