Add MinerU document ingestion support

Normalize uploaded documents into structured markdown, add clearer parser
errors for missing dependencies, and cover the ingestion flow with
backend tests. This also replaces deprecated UTC timestamp helpers in
the touched backend paths so the knowledge pipeline stays warning-free.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

backend/app/services/memory_service.py

@@ -11,6 +11,7 @@ from sqlalchemy import select, desc, func
 from sqlalchemy.ext.asyncio import AsyncSession
 from app.models.memory import MemorySummary, UserMemory
 from app.models.conversation import Conversation, Message
+from app.services.brain_service import BrainService
 from app.services.llm_service import get_llm
 from app.agents.context import get_current_user
 
@@ -235,7 +236,7 @@ async def mark_memory_recalled(db: AsyncSession, memory_id: str):
     if mem:
         mem.is_recalled = True
         mem.recall_count = (mem.recall_count or 0) + 1
-        mem.last_recalled_at = datetime.utcnow()
+        mem.last_recalled_at = datetime.now(UTC)
         await db.commit()
 
 
@@ -271,6 +272,14 @@ async def build_memory_context(
         lines = [f"[对话摘要{i+1}] {s.summary_text}" for i, s in enumerate(recent)]
         parts.append("【之前对话摘要】\n" + "\n".join(lines))
 
+    # 3. 知识大脑（长期项目记忆）
+    brain_memories = await BrainService(db).recall_memories(user_id, current_query, top_k=3)
+    if brain_memories:
+        lines = []
+        for memory in brain_memories:
+            lines.append(f"- {memory.title}: {memory.content}")
+        parts.append("【知识大脑】\n" + "\n".join(lines))
+
     if not parts:
         return ""
     return "\n\n".join(parts)