MC-4636

Build the production integration for the new `what-happened` recency/current-events scout so the daily radio show can use it deterministically, not just as an interactive Hermes skill. Context: - Upstream `last30days` skill is installed at `~/.claude/skills/last30days` and has a silent weekly updater cron. - Hermes adapter skill `what-happened` exists under `~/.hermes/skills/research/what-happened`. - The goal is to use the same behavior for scheduled daily radio/current-events discovery, especially prompts like: `what happened last 1 day on Iran war`. Requirements: 1. Inspect the existing daily radio / briefing curation pipeline and identify the deterministic script/task entrypoint that selects stories before audio generation. 2. Add a `what-happened`/current-events scout step suitable for scheduled runs. Do not rely on the scheduler merely loading a skill; encode the pipeline in script/module logic. 3. Support time windows at minimum: 6h, 24h, 3d, 7d, 30d. Default breaking-news/current-events window should be 24h. 4. Output candidate records with: topic, window, title, why_now, confirmed_new_info, source_quality, source_urls, risk flags, audio_angle, material_update, and suppression_reason where relevant. 5. Enforce radio hygiene from `personalized-audio-briefings`: story ledger dedupe, cooldowns, material-update gate, stale-source labeling, and selected/suppressed rationale. 6. For war/geopolitics/current-events topics, separate confirmed / reported / claimed / social chatter. Viral social claims must not be presented as facts. 7. Add a dry-run mode that prints or saves selected candidates, suppressed repeats, stale/background sources, and rationale without creating audio or posting. 8. Add/update tests or a deterministic fixture for at least one scenario: `last 1 day on Iran war` or equivalent sensitive breaking-news topic. 9. Update docs/runbook/manifest if the production task, script, or scheduled workflow changes. Acceptance criteria: - Dry-run demonstrates a candidate slate and suppression rationale without generating audio. - Existing radio show still runs in preview/dry-run mode. - No production audio/posting side effects during tests unless explicitly enabled. - Implementation is committed/pushed through the mandatory dev-loop. Expected check-in: within 2 hours with either a working dry-run artifact or a concrete blocker.