Sprint 0+1: Package restructure, source tiers, verdicts, multi-variant editions

- New oracle/ package (11 modules) with unified CLI (python -m oracle)
- Source tiers: Tier 1 (arxiv/github/hf), Tier 2 (rss/hn), Tier 3 (reddit)
- Composite verdicts: PUBLISH/WATCH/ARCHIVE/DROP based on signal score + age
- Content-hash dedup: SHA-256[:16] normalized, atomic at insert time
- Multi-variant editions: 4 YAML configs (default/research/devops/brief)
- Variant engine: filter → rank → render (HTML + JSON, themed)
- Per-adapter timeout (10s) + threading fallback
- Consolidated 12 root scripts → thin wrappers + oracle/ package
- Archived stale scripts (_engagement, _live_compare, reddit_proof)
- Updated .gitignore, README.md, schema.sql

variants/research.yaml

@@ -0,0 +1,25 @@
+# Research edition — arXiv + HF papers, high-signal only
+name: "Athena Research"
+description: "Peer-reviewed AI research papers and model releases"
+output: "research/index.html"
+
+filters:
+  verdicts: ["PUBLISH", "WATCH", "ARCHIVE"]
+  sources: ["arxiv", "huggingface"]
+  min_score: 3.0
+  max_age_h: 720
+  max_items: 50
+
+ranking:
+  by: "signal_score"
+  half_life_h: 72
+
+display:
+  top_n: 12
+  show_summary: true
+  show_score: true
+  show_tier: true
+  show_verdict: true
+  theme: "dark"
+  accent: "#a78bfa"
+  logo: "🔬"