docs(comments): drop history narration, keep the why

Rewrite comments that told the story of past rounds instead of the
current code: dated "fixed on / tried on / reverted on" notes, "this used
to...", "stale paragraph removed", "Phase 1/2" progress labels, and the
"ClassifyTile still does not use ClassifyBox" claim (it does now, via
VF_ClassifyBoxWithWarpRetry). Every invariant, unit, sign convention,
trap and measurement that justified the design is kept, rephrased in the
present tense and in its original language (French or English).

Also fix comments that named symbols removed in the dead-code commit
(EvaluateSDF wrapper, FTunnelCoreWorldEvaluation support-floor fields,
BlendChunks -> TransitionBlendChunks) and the stale function name in the
VF_RollStrateParamsDetailed determinism checkf message.

Comment and message text only; no code token changed. Tests untouched.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01URDGLsu7c4nunqNFPJEFPU

Source/VoxelForge/Private/VoxelCaveMorphology.cpp

@@ -51,7 +51,7 @@ namespace
     // Density coordinates and cached cull radii are float-valued.  FVector is double-valued in
     // UE5, so using FVector::DistSquared here needlessly widens every candidate reject.  Keep the
     // cull in the same build-fixed float domain as the density query; the exact field output is
-    // intentionally no longer required to match the pre-optimization double round-trip.
+    // intentionally not required to match a double-precision round-trip.
     FORCEINLINE float VF_FloatDistSquared(
         float X, float Y, float Z, const FVector& Center)
     {
@@ -1356,7 +1356,7 @@ namespace
         }
 
         // Never turn a malformed authored bound into a giant allocation.  A failed build is
-        // deliberately indistinguishable from the old path to the density evaluator.
+        // deliberately indistinguishable from the unindexed linear scan to the density evaluator.
         constexpr int64 MaxBuckets = 1024 * 1024;
         constexpr int64 MaxReferences = 4 * 1024 * 1024;
         TArray<int32> ItemMinX;
@@ -3626,11 +3626,11 @@ namespace
         // Once the lower and upper ends bracket the room, the authored room/floor field is
         // monotone through this narrow vertical landing band for the finite parameter envelope
         // used by the generator.  Find the one-voxel bucket with a logarithmic search, then run
-        // the exact legacy bisection in that bucket.  The latter is kept byte-for-byte in its
-        // arithmetic and iteration count so the endpoint remains identical to the old scan.
+        // the exact bisection in that bucket.  The latter is kept byte-for-byte in its arithmetic
+        // and iteration count so the endpoint is identical to the one-voxel walk below.
         // If the bracket is not monotone (or the local bucket check cannot reproduce the crossing),
-        // fall back to the original one-voxel walk below; malformed/unsupported inputs therefore
-        // retain the old behavior.
+        // fall back to that one-voxel walk; malformed/unsupported inputs therefore keep its
+        // result.
         auto BisectBracket = [&](float BracketLow, float BracketHigh,
                                  float BracketLowValue) -> bool
         {
@@ -7120,14 +7120,14 @@ void VoxelCaveMorphology::BuildChunkCache(
     // (Frontier rooms with no closer candidate in reach fall back to plain NN — a far
     // cluster stays internally chained even when it can't bridge to the origin side.)
     //
-    // bTunnelsFlowTowardOrigin = false (legacy): plain nearest-neighbor pairing. NOTE:
-    // despite what this comment used to claim, an NN-graph is a FOREST of small clusters,
-    // not a connected tree — isolated cave pockets are expected in this mode.
+    // bTunnelsFlowTowardOrigin = false (legacy): plain nearest-neighbor pairing. NOTE: an
+    // NN-graph is a FOREST of small clusters, not a connected tree — isolated cave pockets
+    // are expected in this mode.
     //
     // Selection metric (not the reach filter) penalizes vertical separation via
     // TunnelHorizontalBias, so the GUARANTEED links also prefer walkable slopes —
-    // previously only the random TunnelDensity extras were biased, which is why
-    // backbone tunnels could come out absurdly steep.
+    // biasing only the random TunnelDensity extras lets backbone tunnels come out
+    // absurdly steep.
     TArray<int32, TInlineAllocator<64>> NearestNeighbor;
     NearestNeighbor.SetNumUninitialized(NumRooms);
 
@@ -8266,7 +8266,7 @@ void VoxelCaveMorphology::BuildChunkCache(
         MaxExtent += FloorReliefEnvelope + BlendEnvelope * 3.0f;
 
         // --- PRE-BAKED SHAPE ---
-        // Same hash roll + thresholds + capsule trig the evaluator used to redo PER VOXEL;
+        // Same hash roll + thresholds + capsule trig the evaluator would otherwise redo PER VOXEL;
         // done once here → EvaluateSDFCached just switches on ShapeType. Bit-identical output.
         {
             const uint32 ShapeHash = VoxelHash::Mix(CR.Hash ^ 0xDEADBEEFu);
@@ -8758,7 +8758,7 @@ float VoxelCaveMorphology::EvaluateSDFCached(
     // A room owns every point inside its authored body. The short mouth ownership fade below
     // handles the fillet outside the wall; this broader test is the hard A > B rule for the room
     // interior. The exact raw shape sign was captured during the room SDF pass above, so this
-    // point no longer performs a second spatial-index traversal.
+    // point does not perform a second spatial-index traversal.
     auto EvaluateTunnel = [&](int32 TunnelIdx)
     {
         const FCachedTunnel& Tunnel = TunnelData[TunnelIdx];