refactor(phase0): replace Anthropic SDK with LLMBackend abstraction (default Qwen)

test(phase0): add full mocked test suite for CriticAgent

feat(phase0): add run_critic_gate CLI with rich output and fixture saving

feat(phase0): add CriticEvaluator with rule-based gate logic

feat(phase0): add CriticAgent, CritiqueClaim, CritiqueOutput models