feat(research): add deep-research skill

[Socialpranker]

Summary

• What: Adds a new deep-research skill under the research/ domain — a disciplined, multi-source research workflow (9-phase pipeline: reframe → genre/blocks → plan → capability discovery → parallel-sub-agent search → score & triangulate → synthesize + adversarial pass → verify → refresh targets).
• Why: Fills the heavyweight end of the research domain. The existing research/research skill is a fast router/overview optimized for low decision-risk; deep-research is the rigor-first alternative for high-stakes work — strategy, comparing N options, hypothesis validation — where it adds falsifiable hypotheses, source triangulation (≥3 independent, differently-typed sources per claim), a mandatory adversarial pass, per-source files with verbatim quotes, and a refresh protocol for delta-updates. The SKILL.md explicitly cross-references the router to avoid overlap.

Checklist

• Targets dev branch (not main)
• SKILL.md frontmatter: name + description only
• SKILL.md under 500 lines (87)
• Anti-Patterns section included
• Cross-References included
• No modifications to .codex/, .gemini/, marketplace.json, or docs/
• Clean diff — branched on dev HEAD, single commit, no merge history
• No scripts added (methodology-only skill), so script-specific gates (--help/--json/stdlib) are N/A

Notes

The full, continuously-maintained source catalog (29 channels, 460+ statistical sources, 39 auto-validated APIs, 103 report blocks) lives upstream at https://github.com/Socialpranker/claude-deep-research and is referenced from references/full-catalog.md rather than duplicated here, to keep the skill self-contained while avoiding a large, fast-moving catalog in this repo.