chore: regenerate data files from real sources, remove all fabricated content

Wiped agent_database.db, influencer_database.json, analysis_results.json,
and daily_dispatch_history.json (all fabricated: invented quotes attributed
to real people including Nobel laureates, fake "SUCCESS" publish history)
and regenerated from a clean real scrape across Hacker News, arXiv, BBC
Business, NPR Politics, and BBC World.

Verified: zero "Fresh Scrape"/"Nobel Laureate" strings survive, every
realization keeps a Tech backbone, /api/publish returns DRAFT_READY with
real share-intent links, and the daily scheduler stages a DRAFT_GENERATED
record instead of claiming a fake 10-platform publish.