add domain-discovery and tree-generator meta-skills (v2.1)

Incorporates feedback from TanStack DB testing (discussions #2 and #3):

domain-discovery:
- Cross-domain failure modes with `domains` list for semi-lattice coverage
- Tensions section for design conflicts between domains
- AI-agent-specific failure mode interview questions (Phase 3c)
- Composition discovery from package.json peer deps (Phase 2h)
- Subsystem and reference candidate flagging per domain (Phase 2c)
- Relaxed interview batching for yes/no confirmations
- Failure mode status tracking (active/fixed-but-legacy-risk/removed)
- Strengthened GitHub issue scanning for architectural positions

tree-generator:
- Source repository layout for npm distribution in monorepos
- package.json `files` array guidance for shipping skills
- Framework-integration domain decomposition per adapter package
- Adapter-heavy domain guidance (per-adapter reference files)
- Cross-domain failure modes written into all relevant SKILL files
- Tension notes with cross-references between involved skills
- Failure mode status handling (skip removed, annotate legacy-risk)
- Content-based reference file heuristics (not just length overflow)
- Dense API surface and schema validation reference file triggers

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: KyleAMathews