feat: RenderGraph GPU timestamp profiling

- RG::SetQueryPool 可选注入；Execute 期每 pass 前后各写一枚
  writeTimestamp2(BOTTOM_OF_PIPE)，索引按帧槽位分区 + 执行序排布
- 每帧录制开头 reset 本槽位查询区间（上一轮结果已在 fence 等待后读取）
- Renderer 在 WaitFence 后读 64bit 结果换算毫秒（timestampPeriod 缓存），
  经 GetPassNames/GetPassGpuMs 透出给 ImGui 面板显示
- 每槽位首帧跳过读取（查询尚未 primed，避免 VUID-09401）

Engine/src/RenderGraph/RenderGraph.cpp

@@ -607,14 +607,46 @@ namespace Engine::RG {
         return entry.physical[frameSlot]->view;
     }
 
+    void RenderGraph::SetQueryPool(vk::QueryPool pool, u32 queriesPerFrame)
+    {
+        m_QueryPool = pool;
+        m_QueriesPerFrame = queriesPerFrame;
+    }
+
+    std::vector<std::string> RenderGraph::GetPassNames() const
+    {
+        std::vector<std::string> names;
+        names.reserve(m_ExecutionOrder.size());
+        for (const u32 passIdx : m_ExecutionOrder) {
+            names.push_back(m_Passes[passIdx].name);
+        }
+        return names;
+    }
+
     // ---------- 执行 ----------
 
     void RenderGraph::Execute(const FrameContext& frame)
     {
+        // GPU timestamp：可选。池由 Renderer 创建注入；每帧开头重置本槽位
+        // 区间（上一轮结果已在 fence 等待后被读取），每个 pass 前后各一枚
+        // 时间戳（BOTTOM_OF_PIPE），索引按执行序排布
+        u32 tsPos = 0;
+        if (m_QueryPool) {
+            frame.cmd.resetQueryPool(m_QueryPool,
+                                     frame.frameSlot * m_QueriesPerFrame,
+                                     m_QueriesPerFrame);
+        }
+
         for (const u32 passIdx : m_ExecutionOrder) {
             auto& pass = m_Passes[passIdx];
             if (pass.culled) continue;
 
+            if (m_QueryPool) {
+                frame.cmd.writeTimestamp2(
+                    vk::PipelineStageFlagBits2::eBottomOfPipe,
+                    m_QueryPool, frame.frameSlot * m_QueriesPerFrame + tsPos * 2);
+            }
+
             // barrier：按槽位状态挑选变体，批量提交
             m_ScratchBarriers.clear();
             for (const auto& bp : pass.barriers) {
@@ -652,6 +684,13 @@ namespace Engine::RG {
                 ctx.graph = this;
                 pass.exec(ctx);
             }
+
+            if (m_QueryPool) {
+                frame.cmd.writeTimestamp2(
+                    vk::PipelineStageFlagBits2::eBottomOfPipe,
+                    m_QueryPool, frame.frameSlot * m_QueriesPerFrame + tsPos * 2 + 1);
+            }
+            ++tsPos;
         }
     }