Background BLP texture pre-decoding + deferred WMO normal maps (12x streaming perf)

Move CPU-heavy BLP texture decoding from main thread to background worker
threads for all hot paths: terrain M2 models, WMO doodad M2s, WMO textures,
creature models, and gameobject WMOs. Each renderer (M2, WMO, Character) now
accepts a pre-decoded BLP cache that loadTexture() checks before falling back
to synchronous decode.

Defer WMO normal/height map generation (3 per-pixel passes: luminance, box
blur, Sobel) during terrain streaming finalization — this was the dominant
remaining bottleneck after BLP pre-decoding.

Terrain streaming stalls: 1576ms → 124ms worst case.

include/rendering/vk_context.hpp

@@ -50,9 +50,12 @@ public:
     // Batch upload mode: records multiple upload commands into a single
     // command buffer, then submits with ONE fence wait instead of one per upload.
     void beginUploadBatch();
-    void endUploadBatch();
+    void endUploadBatch();       // Async: submits but does NOT wait for fence
+    void endUploadBatchSync();   // Sync: submits and waits (for load screens)
     bool isInUploadBatch() const { return inUploadBatch_; }
     void deferStagingCleanup(AllocatedBuffer staging);
+    void pollUploadBatches();    // Check completed async uploads, free staging buffers
+    void waitAllUploads();       // Block until all in-flight uploads complete
 
     // Accessors
     VkInstance getInstance() const { return instance; }
@@ -157,6 +160,14 @@ private:
     VkCommandBuffer batchCmd_ = VK_NULL_HANDLE;
     std::vector<AllocatedBuffer> batchStagingBuffers_;
 
+    // Async upload: in-flight batches awaiting GPU completion
+    struct InFlightBatch {
+        VkFence fence = VK_NULL_HANDLE;
+        VkCommandBuffer cmd = VK_NULL_HANDLE;
+        std::vector<AllocatedBuffer> stagingBuffers;
+    };
+    std::vector<InFlightBatch> inFlightBatches_;
+
     // Depth buffer (shared across all framebuffers)
     VkImage depthImage = VK_NULL_HANDLE;
     VkImageView depthImageView = VK_NULL_HANDLE;