feat(asl): Complete Deep Research V2.0 core development

Backend:
- Add SSE streaming client (unifuncsSseClient) replacing async polling
- Add paragraph-based reasoning parser with mergeConsecutiveThinking
- Add requirement expansion service (DeepSeek-V3 PICOS+MeSH)
- Add Word export service with Pandoc, inline hyperlinks, reference link expansion
- Add deep research V2 worker with 2s log flush and Chinese source prompt
- Add 5 curated data sources config (PubMed/ClinicalTrials/Cochrane/CNKI/MedJournals)
- Add 4 API endpoints (generate-requirement/tasks/task-status/export-word)
- Update Prisma schema with 6 new V2.0 fields on AslResearchTask
- Add DB migration for V2.0 fields
- Simplify ASL_DEEP_RESEARCH_EXPANSION prompt (remove strategy section)

Frontend:
- Add waterfall-flow DeepResearchPage (phase 0-4 progressive reveal)
- Add LandingView, SetupPanel, StrategyConfirm, AgentTerminal, ResultsView
- Add react-markdown + remark-gfm for report rendering
- Add custom link component showing visible URLs after references
- Add useDeepResearchTask polling hook
- Add deep research TypeScript types

Tests:
- Add E2E test, smoke test, and Chinese data source test scripts

Docs:
- Update ASL module status (v2.0 - core features complete)
- Update system status (v6.1 - ASL V2.0 milestone)
- Update Unifuncs DeepSearch API guide (v2.0 - SSE mode + Chinese source results)
- Update module auth specification (test script guidelines)
- Update V2.0 development plan

Co-authored-by: Cursor <cursoragent@cursor.com>

backend/src/common/prompt/prompt.fallbacks.ts

@@ -55,7 +55,7 @@ const RVW_FALLBACKS: Record<string, FallbackPrompt> = {
 };
 
 /**
- * ASL 模块兜底 Prompt（预留）
+ * ASL 模块兜底 Prompt
  */
 const ASL_FALLBACKS: Record<string, FallbackPrompt> = {
   ASL_SCREENING: {
@@ -64,6 +64,49 @@ const ASL_FALLBACKS: Record<string, FallbackPrompt> = {
 请根据提供的标准对文献进行筛选，输出JSON格式的结果。`,
     modelConfig: { model: 'deepseek-v3', temperature: 0.2 },
   },
+
+  ASL_DEEP_RESEARCH_EXPANSION: {
+    content: `你是一位经验丰富的医学信息官，擅长将研究者的模糊想法转化为精准的文献检索需求指令。
+
+## 任务
+根据用户输入的粗略研究想法，生成一份简洁的深度文献检索指令书。
+
+## 输出规则
+1. 自然语言风格：像写邮件一样，口语化但专业，方便研究者直接阅读和编辑
+2. PICOS 拆解：明确 Population / Intervention / Comparison / Outcome / Study Design
+3. MeSH 扩展：为关键术语补充 MeSH 同义词（用括号标注英文 MeSH 术语）
+4. 研究设计偏好：若用户未指定，默认优先 RCT、Systematic Review/Meta-Analysis、Cohort Study
+5. 不要输出"检索策略建议"章节
+6. 不要使用 Markdown 加粗标记（即不要用 ** 符号）
+7. 不得自行添加约束：不要擅自限定"仅开放获取"或"仅英文文献"
+
+## 用户输入
+- 研究想法：{{originalQuery}}
+- 选择的数据源：{{targetSources}}
+- 时间范围：{{yearRange}}
+- 目标数量：{{targetCount}}
+
+## 输出格式
+请严格按以下两部分输出，不要添加额外内容：
+
+### Part 1: 自然语言检索指令书
+（简洁的检索需求描述，包含研究背景、PICOS要素、MeSH术语，不要包含检索策略建议）
+
+### Part 2: 结构化摘要（JSON）
+\`\`\`json
+{
+  "objective": "研究目标一句话描述",
+  "population": "研究人群",
+  "intervention": "干预措施（含英文MeSH）",
+  "comparison": "对照组",
+  "outcome": "主要结局指标",
+  "studyDesign": ["RCT", "Meta-analysis"],
+  "meshTerms": ["MeSH术语1", "MeSH术语2"],
+  "condition": "疾病/状况"
+}
+\`\`\``,
+    modelConfig: { model: 'deepseek-v3', temperature: 0.4 },
+  },
 };
 
 /**