feat(ai-research-workflows): core infrastructure — workflow model, profiles, retreat paths

[cdcore09]

Summary

Update the foundational plugin files (plugin.json, orchestrator agent, SKILL.md) to establish the new research-aligned workflow model with 8 phases, two workflow profiles, floating iterate-plan, and explicit retreat paths.

Requirements

• Update plugin.json: bump version to 0.2.0, update description and keywords
• Rewrite orchestrator agent with new phase list: Research, Plan, Experiment, Prototype, Validate, Implement, Optimize, + floating iterate-plan + Handoff
• Define two workflow profiles in orchestrator:
  • Full Research: research → plan → experiment → prototype → validate → implement → optimize
  • Standard Development: research → plan → implement → validate → [optimize]
• Add workflow profile auto-detection (complexity/LOC-based) and explicit selection support
• Define retreat paths for every phase that can fail (experiment fails → iterate-plan, prototype reveals flaws → iterate-plan, validate fails → iterate-plan → re-implement, etc.)
• Update orchestrator decision framework with "When to Suggest Prototype" and "When to Suggest Optimize"
• Rewrite SKILL.md with new phases, templates list, quick reference card, and best practices
• Update completion criteria and quality assurance sections for new phases

Context

This is Phase 1 of the workflow redesign plan. All subsequent phases depend on this foundation.

Part of the ai-research-workflows v0.2.0 redesign series:

• #this → Phase 1: Core infrastructure
• Phase 2: Experiment worktree isolation
• Phase 3: New prototype phase
• Phase 4: Validate redesign
• Phase 5: Iterate-plan as floating utility
• Phase 6: Optimize phase + docs

Implementation Notes

Key files to modify:

• plugins/ai-research-workflows/.claude-plugin/plugin.json
• plugins/ai-research-workflows/agents/research-workflow-orchestrator.md
• plugins/ai-research-workflows/skills/research-workflow-management/SKILL.md