feat(ai-research-workflows): add prototype phase

[cdcore09]

Summary

Create a new prototype command and template. Prototype builds a disposable proof-of-concept in a git worktree to validate the chosen approach end-to-end. The artifact is a findings document — the code is deleted after learnings are captured.

Requirements

• Create commands/prototype.md with YAML frontmatter and full command structure
• Emphasize disposability: findings document is the artifact, NOT the code
• Define clear distinction from experiment: experiment = "which approach?", prototype = "does it work end-to-end?"
• Steps: gather context → define scope → create worktree → build prototype → test against real scenarios → record findings → generate document → clean up worktree
• Create assets/prototype-template.md with sections: Goal, Scope, Approach Used, What Was Built, Test Scenarios, Findings (worked/didn't/surprises/plan accuracy), Recommendations, Retreat Path, References
• Define retreat paths: succeeds → implement, plan flaws → iterate-plan, fundamental failure → experiment or plan
• Allow optional worktree preservation with warning about non-production quality
• Include quality checklist

Context

Phase 3 of the workflow redesign plan. Depends on Phase 1 (orchestrator lists prototype) and Phase 2 (follows experiment's worktree patterns).

This fills the gap between "comparing approaches" (experiment) and "full implementation" (implement). In real research, you build a rough version first, learn from it, then build the real thing.

Implementation Notes

New files to create:

• plugins/ai-research-workflows/commands/prototype.md
• plugins/ai-research-workflows/skills/research-workflow-management/assets/prototype-template.md