feat: add deep-brainstorming skill v2

Research-hardened brainstorming that catches biases AI agents naturally
introduce. 8-phase process with 9 bias types, multi-round research,
adversarial review, and claim provenance tracking.

v2 adds: synthesize-first agent discipline, max 5 questions/agent,
structured doc tool preference, mandatory adversarial review for
Standard intensity, expanded bias taxonomy (6->9 types including
hallucinated evidence, consensus blindspot, phantom requirements),
negative claim verification protocol, and finding engagement depth.

Battle-tested across 14 sessions. 6 quality gate passes.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: belumume