ifcviewer: GPU LOD0/LOD1 selection in compute cull (step 3c)

The compact shader now computes per-instance pixel radius and routes
survivors to LOD1 buckets when the projected sphere falls below the
LOD1 threshold (default 30 px, same as CPU path, tunable via
IFC_LOD1_PX).

Layout expanded from 2 to 4 buckets per mesh:
  [0..M)   fwd_lod0   [M..2M)   fwd_lod1
  [2M..3M) rev_lod0   [3M..4M)  rev_lod1

Two MDIs per model: CCW for [0..2M), CW for [2M..4M).  Per-mesh
has_lod1 flags live in a new gpu_mesh_flags_ssbo (binding 4).

Contribution cull refactored: the compact shader now computes
pixelRadius() once and uses it for both the min_pixel_radius rejection
and LOD routing, matching the CPU path's logic.

Visible-buffer worst case is 2 × total_instances (each LOD bucket
reserves the full fwd/rev capacity per mesh, since LOD selection is
dynamic).

Tri count drops ~60% on the test dataset (53M → 22M) thanks to LOD1
decimated meshes.  FPS recovers from 16 to 36 despite 690k sub_draws
(4M layout).  MDI compaction remains the final perf fix.

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

src/ifcviewer/README.md

@@ -795,14 +795,18 @@ single giant model / <18 cores  CPU BVH trv             Phase 3E GPU cull (plann
 - [x] Event-driven rendering (zero idle CPU/GPU, cull skipped on still frames)
 - [~] **Phase 3E — GPU-side compute-shader culling** (in progress)
       - [x] 3a: `IFC_GPU_CULL=1` drives rendering via compute cull (frustum +
-            contribution, single bucket per mesh).  Correctness matches CPU
-            path; perf regressed — we submit one sub-draw per mesh even
-            when `instanceCount=0`.  Fix is MDI compaction via
-            `glMultiDrawElementsIndirectCount`, deferred to 3a-followup so
-            we don't pull a GL 4.6 entrypoint loader into this commit.
-      - [ ] 3a-followup: compact non-empty commands, use count-buffer MDI
-      - [ ] 3b: fwd/rev reflection bucketing on GPU
-      - [ ] 3c: LOD0/LOD1 selection on GPU
+            contribution).  Perf regressed — submits one sub-draw per mesh
+            even when `instanceCount=0` (CP overhead from empty commands).
+      - [x] 3b: fwd/rev reflection bucketing — compact shader routes by
+            reflected flag into CCW and CW MDI buckets.
+      - [x] 3c: LOD0/LOD1 selection — compact shader computes per-instance
+            pixel radius and routes to LOD1 bucket when below threshold.
+            Per-mesh `has_lod1` flags SSBO.  4 buckets per mesh (fwd/rev ×
+            LOD0/LOD1), 4M commands total, 2 MDIs per model.
       - [ ] 3d: HiZ with same-frame depth pre-pass
+      - [ ] MDI compaction — compact non-empty commands into contiguous
+            buffer, use `glMultiDrawElementsIndirectCount` (GL 4.6 /
+            `ARB_indirect_parameters`).  Deferred until all feature buckets
+            land so we can introduce GL 4.6 loading once, cleanly.
 - [ ] Vulkan/MoltenVK backend for macOS
 - [ ] Embedded Python scripting console