ifcviewer: move scene state into ViewportCore (#84-c)

Move the five scene-state fields that drive per-model GPU upload + the
streaming residency loop into ViewportCore:

  BufferPool      pool_              — vertex+index sub-allocator
  StreamingThread streaming_thread_  — background chunk reader
  std::unordered_map<uint32_t, ModelGpuData> models_gpu_
                                     — per-model state
  uint32_t        next_model_id_     — model-id allocator
  uint32_t        next_object_id_    — globally-unique object-id allocator

ViewportCore.h gains transitive includes for BufferPool / StreamingThread
/ ModelGpuData; ViewportWindow keeps the same names as reference aliases
so existing method bodies that touch them don't have to change.

Same risk profile as #84-a and #84-b: the storage moved but the values
are still set and consumed by the same code paths, so behaviour stays
identical.

Builds: desktop / bonsai / web all green. Tests 100/100.

src/ifcviewer/ViewportCore.h

@@ -35,6 +35,12 @@
 
 #include <webgpu/webgpu.h>
 
+#include <cstdint>
+#include <unordered_map>
+
+#include "BufferPool.h"
+#include "ModelGpuData.h"
+#include "StreamingThread.h"
 #include "ViewportHost.h"
 
 class ViewportCore {
@@ -102,6 +108,27 @@ private:
     // Pick pass. Reuses pipeline_layout_ — same set of bindings as the
     // main pass since the pick fragment also vertex-pulls instance data.
     WGPURenderPipeline  pick_pipeline_ = nullptr;
+
+    // ---- Scene state ---------------------------------------------------------
+    //
+    // Sub-allocator for chunk vertex + index buffers. All per-chunk
+    // pool slices come from here; nothing else uses it. Replaces the
+    // old hand-picked streaming_vram_budget_bytes_ knob entirely.
+    BufferPool pool_;
+
+    // Background worker that does scatter-gather chunk reads off the
+    // render thread. driveStreamingLoads enqueues requests for visible
+    // non-resident chunks and drains completed results into the pool
+    // on subsequent frames.
+    StreamingThread streaming_thread_;
+
+    // Per-model GPU + CPU state, keyed by viewport-assigned model_id.
+    std::unordered_map<uint32_t, ModelGpuData> models_gpu_;
+    uint32_t next_model_id_  = 1;
+    // Globally-unique object_id allocator. Each applyCachedModel rebases
+    // the sidecar's local object_ids by base_object_id_so_far so picks
+    // are unambiguous across models.
+    uint32_t next_object_id_ = 1;
 };
 
 #endif  // VIEWPORTCORE_H

src/ifcviewer/ViewportWindow.cpp

@@ -633,7 +633,12 @@ ViewportWindow::ViewportWindow(QWindow* parent)
       edge_bgl_             (core_.edge_bgl_),
       edge_pipeline_layout_ (core_.edge_pipeline_layout_),
       edge_pipeline_        (core_.edge_pipeline_),
-      pick_pipeline_(core_.pick_pipeline_) {
+      pick_pipeline_(core_.pick_pipeline_),
+      pool_           (core_.pool_),
+      streaming_thread_(core_.streaming_thread_),
+      models_gpu_     (core_.models_gpu_),
+      next_model_id_  (core_.next_model_id_),
+      next_object_id_ (core_.next_object_id_) {
     // wgpu doesn't need a GL context; we just need a real native window
     // whose backing layer matches the GPU API wgpu will drive.
     //

src/ifcviewer/ViewportWindow.h

@@ -984,14 +984,9 @@ public:
     // ceiling that one-WGPUBuffer-per-chunk would otherwise hit. All
     // chunk allocations land here; nothing else uses the pool. Replaces
     // the old hand-picked streaming_vram_budget_bytes_ knob entirely.
-    BufferPool pool_;
+    // Scene-state aliases (storage in core_).
-
+    BufferPool&      pool_;
-    // Background worker that does scatter-gather chunk reads off the
+    StreamingThread& streaming_thread_;
-    // render thread. driveStreamingLoads enqueues requests for visible
-    // non-resident chunks and drains completed results into the pool
-    // on subsequent frames. Kills the 100-300 ms per-frame stutters
-    // that synchronous disk reads caused during orbit.
-    StreamingThread streaming_thread_;
 
     // Per-frame streaming activity, written by driveStreamingLoads,
     // consumed by the benchmark harness to delay the orbit sweep until
@@ -1025,13 +1020,10 @@ private:
     // mirror AppSettings::lod1PixelThreshold() in the GL backend.
     float lod1_pixel_threshold_ = 30.0f;
 
-    // Per-model state, keyed by viewport-assigned model_id.
+    // Per-model state aliases (storage in core_).
-    std::unordered_map<uint32_t, ModelGpuData> models_gpu_;
+    std::unordered_map<uint32_t, ModelGpuData>& models_gpu_;
-    uint32_t next_model_id_  = 1;
+    uint32_t& next_model_id_;
-    // Globally-unique object_id allocator. Each applyCachedModel rebases
+    uint32_t& next_object_id_;
-    // the sidecar's local object_ids by base_object_id_so_far so picks
-    // are unambiguous across models. Selection flags index this range.
-    uint32_t next_object_id_ = 1;
 
     // Sidecar paths queued before init completes.
     std::deque<std::string> pending_sidecars_;