Workflow Engine: JSON DAG 工作流引擎

[DankerMu]

Part of #8

Dependencies: #10 (MVP-β 完成)

Description

自建轻量 JSON DAG 工作流引擎（~500 行），支持拓扑排序、异步并行执行、节点级重试/恢复、SSE 进度报告。预注册 5 个默认 handler。

OpenSpec Spec: `workflow-engine`

Tasks

• 1.1 Implement JSON DAG parser + topological sort
• 1.2 Implement async parallel executor (asyncio.gather per layer)
• 1.3 Implement @engine.register decorator for handler registration
• 1.4 Create workflow_node_runs table
• 1.5 Implement node-level progress reporting via SSE
• 1.6 Implement single-node retry + checkpoint resume
• 1.7 Register 5 default handlers: chapter_summary, kg_extraction, lore_suggestion, consistency_check, export_chapter
• 1.8 Wire chapter_mark_done event → chapter_complete_pipeline workflow
• 1.9 Build workflow progress UI (DAG visualization + node status)
• 1.10 Write tests for DAG parsing, parallel execution, retry, resume

Acceptance Criteria

• JSON DAG 支持定义节点依赖和并行执行
• 节点失败可单独重试，不影响已完成节点
• chapter_mark_done 自动触发完整 pipeline
• 前端可视化 DAG 执行进度