fix: address cubic AI review feedback

- Use accumulator with extend() instead of list concatenation to avoid
  quadratic CPU/memory churn on large paginated listings
- Absorb final aggregated tools result as complete to prune stale session
  state (x-mcp-header maps and output schemas)

Fixes cubic review comments on PR #3156

Signed-off-by: dongjiang <dongjiang1989@126.com>

Author: dongjiang1989

src/mcp/client/client.py

@@ -992,17 +992,21 @@ async def _drain_all_pages(
         pages = 1
         max_pages = self.list_max_pages
 
+        # Use an accumulator to avoid quadratic list copying
+        accumulator = get_items(first)
+
         while cursor is not None:
             if cursor in seen:
                 raise CursorCycleError(method, cursor)
             seen.add(cursor)
             if max_pages > 0 and pages >= max_pages:
                 raise PaginationExceededError(method, max_pages)
             page = await fetch_page(cursor)
-            set_items(first, get_items(first) + get_items(page))
+            accumulator.extend(get_items(page))
             cursor = page.next_cursor
             pages += 1
 
+        set_items(first, accumulator)
         # Strip the terminal cursor from the aggregated result.
         first.next_cursor = None
         return first
@@ -1014,12 +1018,14 @@ async def list_all_tools(self) -> ListToolsResult:
             PaginationExceededError: The walk exceeded ``list_max_pages``.
             CursorCycleError: The server returned a repeated cursor.
         """
-        return await self._drain_all_pages(
+        result = await self._drain_all_pages(
             "tools/list",
             fetch_page=lambda c: self.list_tools(cursor=c, cache_mode="bypass"),
             get_items=lambda r: list(r.tools),
             set_items=lambda r, items: setattr(r, "tools", items),  # noqa: B010
         )
+        # Absorb the final aggregated result as complete to prune stale session state
+        return self.session._absorb_tool_listing(result, complete=True)  # pyright: ignore[reportPrivateUsage]
 
     async def list_all_resources(self) -> ListResourcesResult:
         """Fetch all resources from the server, draining pagination automatically.