feat: TAA 时域抗锯齿（velocity + 邻域 clamp + 子像素抖动）

- Velocity buffer：Forward 第二颜色附件，vertex 计算 clip 空间运动向量
  （UBO 尾部追加 projNoJitter/prevView/prevProj；上一帧实例矩阵 SSBO
  Set0 binding2，与当前实例数组平行对齐——转动的物体速度正确不鬼影）
- 子像素抖动：SceneRenderer Halton(2,3)×8 序列，抖动只进光栅化投影
- TAA resolve pass：velocity 反推历史位置 + 3×3 邻域 clamp 去鬼影 +
  mix(cur,hist,0.9) 时域累积；TAAColor 自引用历史
- RG 历史读补 post-barrier：采样后恢复镜像 resting layout（TAA 的历史
  是 ColorAttachment layout，不恢复则下帧 entry oldLayout 脱节）
- 管线：Forward(HDR+Velocity) → TAA(TAAColor) → Bloom/Tonemap 改源 TAAColor
- 编译零错误、validation 零警告、8 秒冒烟通过（10 pass 全活跃）

Engine/src/Renderer/Renderer.cpp

@@ -14,6 +14,7 @@
 #include <Engine/Renderer/Passes/RenderPass.h>
 #include <Engine/Renderer/Passes/ShadowPass.h>
 #include <Engine/Renderer/Passes/ForwardPass.h>
+#include <Engine/Renderer/Passes/TAAPass.h>
 #include <Engine/Renderer/Passes/BloomPass.h>
 #include <Engine/Renderer/Passes/TonemapPass.h>
 #include <Engine/Renderer/Passes/TemporalDiffPass.h>
@@ -48,19 +49,23 @@ namespace Engine
         m_DescriptorManager = CreateScope<DescriptorManager>(*m_Device);
         m_ShaderManager = CreateScope<ShaderManager>(*m_Device);
 
-        // 2. Per-frame ring：UBO + GPUDriven 实例 SSBO（同一 ring 类，
-        //    SSBO 只是 usage 不同；正确性保证一致）
+        // 2. Per-frame ring：UBO + GPUDriven 实例 SSBO（当前/上一帧，
+        //    同一 ring 类，SSBO 只是 usage 不同；正确性保证一致）
         m_UniformBuffer = CreateScope<UniformBuffer>(
             *m_Device, *m_Allocator, sizeof(UniformBufferObject), kMaxFramesInFlight);
         m_ObjectBuffer = CreateScope<UniformBuffer>(
             *m_Device, *m_Allocator,
             sizeof(glm::mat4) * kMaxInstances, kMaxFramesInFlight,
             vk::BufferUsageFlagBits::eStorageBuffer);
+        m_PrevObjectBuffer = CreateScope<UniformBuffer>(
+            *m_Device, *m_Allocator,
+            sizeof(glm::mat4) * kMaxInstances, kMaxFramesInFlight,
+            vk::BufferUsageFlagBits::eStorageBuffer);
 
         // 3. pass 装配体：布局/shader/采样器/per-slot 描述符
         m_SceneGpu = CreateScope<Passes::SceneGpu>(
             *m_Device, *m_DescriptorManager, *m_ShaderManager,
-            *m_UniformBuffer, *m_ObjectBuffer);
+            *m_UniformBuffer, *m_ObjectBuffer, *m_PrevObjectBuffer);
         m_FullscreenGpu = CreateScope<Passes::FullscreenGpu>(
             *m_Device, *m_DescriptorManager, *m_ShaderManager);
 
@@ -183,12 +188,22 @@ namespace Engine
             "ShadowMap",
             RG::TextureDesc{vk::Format::eD32Sfloat, kShadowMapSize, kShadowMapSize});
 
-        // HDR 颜色图：Forward 的输出（线性光照值），Tonemap 显式完成
-        // 显示变换后写入 Backbuffer（对齐 UE/Unity 管线）
+        // HDR 颜色图：Forward 的输出（线性光照值 + 抖动），TAA 显式完成
+        // 时域累积后写入 TAAColor（对齐 UE/Unity 管线）
         m_HDRColor = m_RenderGraph->CreateTexture(
             "HDRColor",
             RG::TextureDesc{vk::Format::eR16G16B16A16Sfloat});
 
+        // velocity：clip 空间运动向量（TAA 反推历史像素位置）
+        m_Velocity = m_RenderGraph->CreateTexture(
+            "Velocity",
+            RG::TextureDesc{vk::Format::eR16G16B16A16Sfloat});
+
+        // TAA 累积输出：既是 resolve 结果也是下一帧的历史（自引用）
+        m_TAAColor = m_RenderGraph->CreateTexture(
+            "TAAColor",
+            RG::TextureDesc{vk::Format::eR16G16B16A16Sfloat});
+
         // Bloom ping-pong：半分辨率（sizeScale 相对目标尺寸，resize 自动跟随）
         m_BloomA = m_RenderGraph->CreateTexture(
             "BloomA",
@@ -209,13 +224,16 @@ namespace Engine
         m_Passes.push_back(CreateScope<Passes::ShadowPass>(
             *m_SceneGpu, frame, m_ShadowMap));
         m_Passes.push_back(CreateScope<Passes::ForwardPass>(
-            *m_SceneGpu, frame, m_HDRColor, m_Depth, m_ShadowMap));
+            *m_SceneGpu, frame, m_HDRColor, m_Velocity, m_Depth, m_ShadowMap));
+        m_Passes.push_back(CreateScope<Passes::TAAPass>(
+            *m_FullscreenGpu, /*historyAlpha*/ 0.9f, m_HDRColor, m_Velocity,
+            m_TAAColor));
         m_Passes.push_back(CreateScope<Passes::TemporalDiffPass>(
             *m_FullscreenGpu, /*gain*/ 8.0f, m_HDRColor, m_TemporalDiff));
         m_Passes.push_back(CreateScope<Passes::BloomPass>(
-            *m_FullscreenGpu, *m_PostParams, m_HDRColor, m_BloomA, m_BloomB));
+            *m_FullscreenGpu, *m_PostParams, m_TAAColor, m_BloomA, m_BloomB));
         m_Passes.push_back(CreateScope<Passes::TonemapPass>(
-            *m_FullscreenGpu, *m_PostParams, m_HDRColor, m_BloomA,
+            *m_FullscreenGpu, *m_PostParams, m_TAAColor, m_BloomA,
             m_TemporalDiff, m_Backbuffer));
         m_Passes.push_back(CreateScope<Passes::ImGuiPass>(
             &m_ImGuiLayer, m_Backbuffer));
@@ -295,9 +313,7 @@ namespace Engine
         const auto& vkDevice = m_Device->GetDevice();
 
         // 1. 场景 → 当帧渲染数据（相机/灯光/双视图剔除；纯 CPU，不碰 GPU）
-        const f32 aspect = static_cast<f32>(m_Extent.width) /
-                           static_cast<f32>(m_Extent.height);
-        if (!m_SceneRenderer->BuildFrame(scene, assets, aspect)) {
+        if (!m_SceneRenderer->BuildFrame(scene, assets, m_Extent)) {
             return;   // 无 primary 相机：跳帧
         }
         const FrameSceneData& frame = m_SceneRenderer->Frame();
@@ -311,6 +327,10 @@ namespace Engine
         if (!frame.instanceModels.empty()) {
             m_ObjectBuffer->Update(frameSlot, frame.instanceModels.data(),
                                    frame.instanceModels.size() * sizeof(glm::mat4));
+            // 上一帧姿态与当前帧平行对齐（velocity 顶点着色器按下标取）
+            m_PrevObjectBuffer->Update(frameSlot, frame.prevInstanceModels.data(),
+                                       frame.prevInstanceModels.size() *
+                                           sizeof(glm::mat4));
         }
 
         // 3. acquire swapchain image