Evaluate an Agent System Design with a Rubric

Implement evaluate_design(proposal, rubric) that scores an agent system design against a structured rubric. proposal is a dict mapping category names to dicts of design decisions: {"orchestration": {"framework": str, "loop_type": str, ...}, "tools": {...}, ...} rubric is a dict mapping category names to lists of criteria. Each criterion is a dict: - key (str): the design decision key to check in proposal[category]. - label (str): human-readable criterion name. - max_points (int): maximum score if present. - weight (float, optional): importance multiplier (default 1.0). Requirements: - For each category in rubric, check that the category exists in proposal and each criterion key is present. - Score each present criterion at its max_points (weighted). Missing criteria score 0. - Identify missing failure handling: if a category exists but lacks any "fallback", "retry", or "error" key, note it. - Prioritize feedback: missing components first, then underdeveloped categories (< 50% of max). - Return {"total_score": float, "max_score": float, "percentage": float, "feedback": list[str], "details": dict}.