feat: processing flow enhancement + responsive UI

Phase 2 - Processing flow:
- Multi-task monitoring: store supports concurrent task tracking
- Task retry: POST /api/tasks/{id}/retry re-runs failed tasks
- Dashboard multi-task cards with progress, error details, retry/dismiss
- Log panel expanded from 10 to 50 lines with "view all" link

Phase 3 - UI/UX:
- Mobile sidebar drawer (< 768px) with hamburger menu
- Layout responsive styles (768px, 480px breakpoints)
- Tasks/Logs pages responsive (stat cards, filters, columns)
- File views responsive (header wrap, button sizing)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

web/backend/routers/tasks.py

@@ -121,7 +121,34 @@ async def retry_task(
     """Retry a failed task by re-invoking its processing endpoint.
 
     Only tasks with status ``failed`` may be retried.
+    For in-memory tasks with metadata, the original endpoint and request body
+    are used to faithfully reproduce the original call.  For historical DB-only
+    tasks, the endpoint is looked up from ``_RETRY_ROUTE_MAP`` by task name.
     """
+    tm = request.state.task_manager
+
+    # --- Strategy 1: in-memory task with metadata ---
+    new_task = tm.retry_task(task_id)
+    if new_task is not None:
+        meta = new_task.metadata or {}
+        endpoint = meta.get("endpoint")
+        body = meta.get("body", {})
+        if endpoint:
+            base_url = f"http://{request.url.hostname}:{request.url.port}"
+            url = f"{base_url}{endpoint}"
+            auth_header = request.headers.get("authorization")
+            headers: dict[str, str] = {}
+            if auth_header:
+                headers["authorization"] = auth_header
+            async with httpx.AsyncClient() as client:
+                resp = await client.post(url, json=body, headers=headers)
+            return {"task_id": new_task.id, "status": "retried", "original_response": resp.json()}
+
+        # Metadata present but no endpoint — fall through to DB strategy
+        # (the new task was already created; caller can track it)
+        return {"task_id": new_task.id, "status": "retried"}
+
+    # --- Strategy 2: DB-only historical task (no in-memory record) ---
     loop = asyncio.get_event_loop()
     task = await loop.run_in_executor(
         None, lambda: db_schema.query_task_by_id(task_id),
@@ -142,18 +169,18 @@ async def retry_task(
             detail=f"未知的任务类型: {task_name}",
         )
 
-    # Build the internal URL to the processing endpoint.
+    # Create a new in-memory task to track the retry.
+    new_task = tm.create_task(task_name)
+
     base_url = f"http://{request.url.hostname}:{request.url.port}"
     url = f"{base_url}{endpoint}"
 
-    # Forward the Authorization header so the processing endpoint can
-    # authenticate the request.
     auth_header = request.headers.get("authorization")
-    headers: dict[str, str] = {}
+    headers = {}
     if auth_header:
         headers["authorization"] = auth_header
 
     async with httpx.AsyncClient() as client:
         resp = await client.post(url, headers=headers)
 
-    return resp.json()
+    return {"task_id": new_task.id, "status": "retried", "original_response": resp.json()}