perf(gen): a thrashing one-slot cache cost 18.7x -- and the world is unchanged

TunnelNetwork density evaluation was 45x more expensive than Maze for identical
work: 2,097,152 voxels, 64 tiles, same mesher, same machine. Profiling killed
every hypothesis I brought to it -- box classification skipping 0/64 tiles was
true but irrelevant (export bypasses classification), near-surface early-outs
fired at 49.7%, the room cache built 340 entries for 2.74M samples, and the
per-room overrides resolved once per sample rather than once per modifier.

The cause was the wrapper, not the stack. A mesher tile samples an EXPANDED grid
(-1..Cells+1), so one worker crosses several exact chunk coordinates while
meshing a single tile. The generator's cache held only the last coordinate, so
every boundary crossing rebuilt the resolved parameters, the whole 22-operator
stack and the tunnel-core cache. On one representative tile: 3,675 refetches,
38.0s. The entire operator stack was 18.5s of a 4,272s total -- 0.43%.

Replaced with a 128-slot direct-mapped thread_local cache keyed on owner, chunk,
layout version, params fingerprint and generator type. Collisions force a safe
rebuild, never a stale read. No shared mutable state, no order-dependent
reduction.

  TunnelNetwork  344.5s / 164.267 us/vox  ->  18.4s / 8.763 us/vox   (18.7x)
  Maze             7.4s /   3.535 us/vox  ->   1.08s / 0.513 us/vox  ( 6.9x)
  same tile        38.0s                  ->   0.51s ; refetches 3675 -> 27

Output is BYTE-IDENTICAL, before vs after and across repeats, both archetypes:
TunnelNetwork 50d6b4d0..f23126, Maze 364f7489..2fbe99. The permission to change
the field was not needed.

Also in this commit, from the two preceding tasks:

- Explorer renders from the MESH instead of raymarching. It meshes the region
  once through the canonical UVoxelMarchingCubesMesher, then CPU-rasterises via
  a BVH. A viewpoint went from ~16 minutes to 0.155s; 8 views now cost 1.25s of
  drawing. No render pixel samples the density field.
- One sampled grid and player-fit mask per run, shared by render/walk/export,
  instead of the same 368s walk computed twice.
- A wall-clock budget (25 min default, hard 30 max) that stops cleanly and
  records truncated/truncated_during/completed_modes rather than overshooting.
- ParallelFor across the measurement sampling; the suite was ~10% of a 16-thread
  machine.
- -surfaceroughness= on the commandlet, and a surface-relief metric in export
  mode (edge lengths, dihedral distribution, face-vs-smooth normal divergence)
  so "does this look like rock" is a number, not a screenshot argument.
- Smooth vertex normals in the rasteriser: FVoxelMeshData::Normals is the
  density-gradient normal the engine shades with, and the raster path was
  recomputing a face normal over it.
- ScaleDiagnosis: baked candidate indices are now derived from the current
  corpus, so a legitimate field change stops breaking the test.

NOT verified, and not claimed:
- TunnelNetworkSpineEquivalence exceeded the 30-minute cap and was stopped.
  Inconclusive, not passing. The identical hashes are strong evidence the
  op-stack/legacy equivalence holds, but the test has not confirmed it.
- CrossPlatformDigest passes but reports 2 samples within 1e-6 of the
  isosurface; FMath::Sin/Cos leaves Windows/Linux equivalence unproven.
  Pre-existing, not caused by this change.
- The cache's memory footprint (128 entries x ~16 workers) is unmeasured.

What the roughness sweep settled, for the record: SurfaceRoughness 0 -> 8 moves
cost by 0.3% (163.2 / 163.2 / 163.7 us), so roughness was never the cost. It
does reach the surface (dihedral p50 1.1 -> 2.2 -> 5.9), which kills my claim
that the density-channel operator contributed nothing. TunnelNetwork's hard
seams are present at roughness 0 (4.4% of edges past 40 deg vs Maze's 1.0%), so
they are structural -- in how primitives are unioned -- not a roughness
deficiency. More roughness makes them worse, not better.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012aheo2crrZmHuYtPpdY2Cb

CODEMAP.md

@@ -397,20 +397,23 @@ No UObject state, cache, actor, world, or PIE is required.
 | Symbol | Role |
 |--------|------|
 | `FVoxelStrateMeasureSettings` / `FVoxelStrateMetrics` | Plain settings/result structs for bounded strate sampling and derived measurements; callers may override the interior margin (including zero to sample the seal) or fit the XY window to two points with a margin, and results identify the resolved window/dimensions/Z range, deterministic air-component facts, walkable floor-area columns/fraction, projected surface components/share, and median clearance. |
-| `FVoxelStrateSampleGrid` | Optional one-pass `density > 0` `Air` polarity plus exact scalar `Density` capture exported by measurement functions; bounded by `MaxCells`, consumed by editor/automation previews, never retained on ordinary metrics calls. |
+| `FVoxelStrateSampleGrid` | Optional one-pass `density > 0` `Air` polarity plus exact scalar `Density` capture; player-fit walks may also export the exact `PlayerFitMask` and count. Bounded by `MaxCells`, shared by explorer render/walk/export, and never retained on ordinary metrics calls. |
 | `VF_MeasureStrate` | One-grid/one-flood-fill strate metrics; resolves either the legacy centered square or the two-point fitted AABB, and refuses invalid bounds or a grid over `MaxCells`. |
 | `VF_AreConnected` | Coarse 6-connected BFS plus deterministic blocked-edge retries, with a full-resolution air recheck for every candidate route and explicit endpoint-solid, out-of-window, `NotConnectedAtThisResolution`, `CoarseLiedBudgetExhausted` (unknown), and connected outcomes; diagnostics include endpoint component facts and the retry count from that same grid. |
 | `VF_DiagnoseConnectivity` | The same query plus each mouth's air-component size/share and exact geometric distance to the nearest cell of the other component; proximity is measured in coarse-cell coordinates and does not claim a route through solid. |
-| `FVoxelPlayerFitWalkReport` / `VF_MeasurePlayerFitWalkWithSampler` | Editor-only experiential instrument over the existing exact `VF_` capsule/floor-fit mask: deterministic graph-walk distance, dead ends, reachable fit volume, and a labelled narrow-gap proxy; no second movement model. |
+| `FVoxelPlayerFitWalkReport` / `VF_MeasurePlayerFitWalkWithSampler` | Editor-only experiential instrument over the existing exact `VF_` capsule/floor-fit mask: deterministic graph-walk distance, dead ends, reachable fit volume, and a labelled narrow-gap proxy; optionally exports the one sampled grid/mask for other explorer consumers; no second movement model. |
 
 `VoxelForgeExploreCommandlet` builds a transient, fixed-definition world and calls the production
-`UVoxelGenerator::GetDensityAt` and `UVoxelMarchingCubesMesher::GenerateMesh` paths. `render` uses
-fixed density samples (default 0.25 voxel) with bisection at sign crossings instead of sphere tracing;
-`walk` uses `VF_MeasurePlayerFitWalkWithSampler`; `export` calls the canonical 32³ mesher in a
-deterministic tile grid (four tiles per axis for the 128³ default), streams only its returned tile
-arrays into metre-space OBJ plus `manifest.json` (canonical UVs remain voxel-space), and never retains the full region mesh. Output JSON/manifest are emitted by fixed-order writers and compared with a second
-serialization before it is written. The commandlet keeps `WorldRadiusVoxels=0`, does not touch an
-authored asset or diff layer, and refuses bounded-grid/export requests before allocating them.
+`UVoxelGenerator::GetDensityAt` and `UVoxelMarchingCubesMesher::GenerateMesh` paths. `walk` (or the
+private seed walk used by render-only runs) captures one exact sampled grid and player-fit mask;
+render and export reuse that hand-off. `render` calls the canonical 32³ mesher in a deterministic
+tile grid (four tiles per axis for the 128³ default), merges the region once, builds a deterministic
+CPU BVH, and rasterises eight viewpoints from the mesh without per-pixel density calls. `export`
+uses that same canonical aggregate for metre-space OBJ plus `manifest.json` (canonical UVs remain
+voxel-space). Output JSON/manifest are emitted by fixed-order writers and compared with a second
+serialization before they are written. The commandlet keeps `WorldRadiusVoxels=0`, does not touch an
+authored asset or diff layer, and enforces the default 25-minute / maximum 30-minute wall-clock
+budget, reporting partial artifacts and completed modes on truncation.
 
 ### 3.7 Cave morphology (SDF rooms/tunnels) — `Public/VoxelCaveMorphology.h` + `.cpp`
 Header is rich with inline docs. Two namespaces + a per-chunk cache system.