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/Passes/ForwardPass.cpp

@@ -8,6 +8,7 @@
 #include <Engine/Renderer/Material.h>
 
 #include <array>
+#include <span>
 
 namespace Engine::Passes
 {
@@ -20,12 +21,21 @@ namespace Engine::Passes
             clearValue.color = vk::ClearColorValue(
                 std::array<float, 4>{0.1f, 0.1f, 0.1f, 0.0f});
 
-            vk::RenderingAttachmentInfo colorAttachment;
-            colorAttachment.setImageView(ctx.View(m_HDRColor))
-                           .setImageLayout(vk::ImageLayout::eColorAttachmentOptimal)
-                           .setLoadOp(vk::AttachmentLoadOp::eClear)
-                           .setStoreOp(vk::AttachmentStoreOp::eStore)
-                           .setClearValue(clearValue);
+            vk::RenderingAttachmentInfo colorAttachments[2];
+            colorAttachments[0].setImageView(ctx.View(m_HDRColor))
+                               .setImageLayout(vk::ImageLayout::eColorAttachmentOptimal)
+                               .setLoadOp(vk::AttachmentLoadOp::eClear)
+                               .setStoreOp(vk::AttachmentStoreOp::eStore)
+                               .setClearValue(clearValue);
+
+            // velocity 附件：清 0 = 静止（背景无几何 → 历史原位采样）
+            vk::ClearValue velocityClear;
+            velocityClear.color = vk::ClearColorValue(std::array<float, 4>{0.f, 0.f, 0.f, 0.f});
+            colorAttachments[1].setImageView(ctx.View(m_Velocity))
+                               .setImageLayout(vk::ImageLayout::eColorAttachmentOptimal)
+                               .setLoadOp(vk::AttachmentLoadOp::eClear)
+                               .setStoreOp(vk::AttachmentStoreOp::eStore)
+                               .setClearValue(velocityClear);
 
             vk::RenderingAttachmentInfo depthAttachment;
             depthAttachment.setImageView(ctx.View(m_Depth))
@@ -35,10 +45,13 @@ namespace Engine::Passes
                            .setClearValue(vk::ClearValue(
                                vk::ClearDepthStencilValue(1.0f, 0)));
 
+            std::array<vk::RenderingAttachmentInfo, 2> colorArray = {
+                colorAttachments[0], colorAttachments[1]};
+
             vk::RenderingInfo renderingInfo;
             renderingInfo.setRenderArea({{0, 0}, ctx.extent})
                          .setLayerCount(1)
-                         .setColorAttachments(colorAttachment)
+                         .setColorAttachments(colorArray)
                          .setPDepthAttachment(&depthAttachment);
 
             cmd.beginRendering(renderingInfo);
@@ -95,8 +108,6 @@ namespace Engine::Passes
             cmd.setScissorWithCount(scissor);
 
             vk::Bool32 blendEnable = VK_FALSE;
-            cmd.setColorBlendEnableEXT(0, blendEnable);
-
             vk::ColorBlendEquationEXT blendEquation;
             blendEquation.setSrcColorBlendFactor(vk::BlendFactor::eOne)
                          .setDstColorBlendFactor(vk::BlendFactor::eZero)
@@ -104,14 +115,20 @@ namespace Engine::Passes
                          .setSrcAlphaBlendFactor(vk::BlendFactor::eOne)
                          .setDstAlphaBlendFactor(vk::BlendFactor::eZero)
                          .setAlphaBlendOp(vk::BlendOp::eAdd);
-            cmd.setColorBlendEquationEXT(0, blendEquation);
 
             vk::ColorComponentFlags colorWriteMask =
                 vk::ColorComponentFlagBits::eR |
                 vk::ColorComponentFlagBits::eG |
                 vk::ColorComponentFlagBits::eB |
                 vk::ColorComponentFlagBits::eA;
-            cmd.setColorWriteMaskEXT(0, colorWriteMask);
+
+            // 两个颜色附件（HDR + velocity）都是无混合直写
+            cmd.setColorBlendEnableEXT(0,
+                std::array<vk::Bool32, 2>{blendEnable, blendEnable});
+            cmd.setColorBlendEquationEXT(0,
+                std::array<vk::ColorBlendEquationEXT, 2>{blendEquation, blendEquation});
+            cmd.setColorWriteMaskEXT(0,
+                std::array<vk::ColorComponentFlags, 2>{colorWriteMask, colorWriteMask});
 
             // ---- 绘制：GPUDriven 按 mesh 分 batch（相机视锥剔除后）----
             // 实例矩阵已由 Renderer 写入 ObjectBuffer（per-slot SSBO）；
@@ -127,6 +144,7 @@ namespace Engine::Passes
             cmd.endRendering();
         })
         .Write(m_HDRColor, RG::ResourceUse::ColorAttachment)
+        .Write(m_Velocity, RG::ResourceUse::ColorAttachment)
         .Write(m_Depth, RG::ResourceUse::DepthAttachment)
         .Read(m_ShadowMap, RG::ResourceUse::ShaderRead,
               vk::ShaderStageFlagBits::eFragment);