feat(gpu): migrar a SDL3_GPU amb 2-pass rendering i post-processat

- Infraestructura GPU: GpuContext, GpuPipeline, GpuSpriteBatch, GpuTexture
- Engine::render() migrat a 2-pass: sprites → offscreen R8G8B8A8 → swapchain + vignette
- UI/text via software renderer (SDL3_ttf) + upload com a textura overlay GPU
- CMakeLists.txt actualitzat per incloure subsistema gpu/

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

source/gpu/gpu_sprite_batch.hpp

@@ -0,0 +1,86 @@
+#pragma once
+
+#include <SDL3/SDL_gpu.h>
+#include <vector>
+#include <cstdint>
+
+// ---------------------------------------------------------------------------
+// GpuVertex — 8-float vertex layout sent to the GPU.
+// Position is in NDC (pre-transformed on CPU), UV in [0,1], color in [0,1].
+// ---------------------------------------------------------------------------
+struct GpuVertex {
+    float x, y;        // NDC position  (−1..1)
+    float u, v;        // Texture coords (0..1)
+    float r, g, b, a;  // RGBA color     (0..1)
+};
+
+// ============================================================================
+// GpuSpriteBatch — Accumulates sprite quads, uploads them in one copy pass.
+//
+// Usage per frame:
+//   batch.beginFrame();
+//   batch.addBackground(...);       // Must be first (bg indices = [0..5])
+//   batch.addSprite(...) × N;
+//   batch.uploadBatch(device, cmd); // Copy pass
+//   // Then in render pass: bind buffers, draw bg with white tex, draw sprites.
+// ============================================================================
+class GpuSpriteBatch {
+public:
+    // Maximum sprites (background + UI overlay each count as one sprite)
+    static constexpr int MAX_SPRITES = 65536;
+
+    bool init(SDL_GPUDevice* device);
+    void destroy(SDL_GPUDevice* device);
+
+    void beginFrame();
+
+    // Add the full-screen background gradient quad.
+    // top_* and bot_* are RGB in [0,1].
+    void addBackground(float screen_w, float screen_h,
+                       float top_r, float top_g, float top_b,
+                       float bot_r, float bot_g, float bot_b);
+
+    // Add a sprite quad (pixel coordinates).
+    // scale: uniform scale around the quad centre.
+    void addSprite(float x, float y, float w, float h,
+                   float r, float g, float b, float a,
+                   float scale,
+                   float screen_w, float screen_h);
+
+    // Add a full-screen overlay quad (e.g. UI surface, NDC −1..1).
+    void addFullscreenOverlay();
+
+    // Upload CPU vectors to GPU buffers via a copy pass.
+    // Returns false if the batch is empty.
+    bool uploadBatch(SDL_GPUDevice* device, SDL_GPUCommandBuffer* cmd_buf);
+
+    SDL_GPUBuffer* vertexBuffer()  const { return vertex_buf_; }
+    SDL_GPUBuffer* indexBuffer()   const { return index_buf_; }
+    int bgIndexCount()             const { return bg_index_count_; }
+    int overlayIndexOffset()       const { return overlay_index_offset_; }
+    int overlayIndexCount()        const { return overlay_index_count_; }
+    int spriteIndexOffset()        const { return sprite_index_offset_; }
+    int spriteIndexCount()         const { return sprite_index_count_; }
+    bool isEmpty()                 const { return vertices_.empty(); }
+
+private:
+    void toNDC(float px, float py, float screen_w, float screen_h,
+               float& ndx, float& ndy) const;
+    void pushQuad(float ndx0, float ndy0, float ndx1, float ndy1,
+                  float u0, float v0, float u1, float v1,
+                  float r, float g, float b, float a);
+
+    std::vector<GpuVertex> vertices_;
+    std::vector<uint32_t>  indices_;
+
+    SDL_GPUBuffer*         vertex_buf_      = nullptr;
+    SDL_GPUBuffer*         index_buf_       = nullptr;
+    SDL_GPUTransferBuffer* vertex_transfer_ = nullptr;
+    SDL_GPUTransferBuffer* index_transfer_  = nullptr;
+
+    int bg_index_count_       = 0;
+    int sprite_index_offset_  = 0;
+    int sprite_index_count_   = 0;
+    int overlay_index_offset_ = 0;
+    int overlay_index_count_  = 0;
+};