refactor(physics): delete legacy PhysicsEngine.Resolve/ResolvePlacement/HasCellSurface (C5a, AP-1/AD-1)

Member-wise deletion of the three legacy resolver members named in
docs/research/2026-08-05-c5a-contract.md: PhysicsEngine.Resolve,
PhysicsEngine.HasCellSurface, and PhysicsEngine.ResolvePlacement. An
exhaustive receiver census over src/ found zero production callers of any
of the three — every production placement writer already reaches the
canonical PhysicsEngine.SetPosition transaction exclusively through
RuntimeSetPositionState (three call sites total). The deletion is purely
member-wise: IsSpawnCellReady and AdjustPosition, which shared the same
source region as the deleted members, are preserved byte-identical — every
remaining production caller of either (including PhysicsCameraCollisionProbe,
AdjustPosition's sole surviving production caller) is unaffected.

Companion changes:
- PlayerMovementController's 3-argument SetPosition test overload is renamed
  to SeedPlacementForTest (internal) and CommitPreparedPosition is deleted;
  83 call sites across 19 test files were mechanically renamed to match.
- Seven pinned test dispositions from the contract are executed:
  3.1 (PhysicsEngineTests.cs: 11 legacy-resolver tests deleted, 6
  ResolveWithTransition tests kept), 3.2/3.3/3.4 (re-point to canonical
  SetPosition, with TransitionScratchDifferentialTests.cs additionally
  gaining positive IsCommitted assertions after each bitwise comparison so
  the differential proves a placement actually committed, not just that two
  possibly-uncommitted results match), 3.5 (Runtime rename), and 3.6
  (PlayerMovementPlacementTransactionTests.cs rewritten — its xmldoc now
  states plainly that the render-root publish moved to
  RuntimeSetPositionState.cs, but the sticky-release relocation claim was
  false and is retracted; this disposition's coverage loss is the sticky
  release path, not silently absorbed elsewhere).
- Stale `PhysicsEngine.Resolve`/`Resolve` doc citations in CellTransit.cs,
  PlayerMovementController.cs, and HeadlessSessionWorldProjection.cs are
  corrected to name the surviving canonical entry points by symbol
  (SetPosition, AdjustSetPosition/AdjustPosition, ResolveWithTransition)
  rather than fragile line numbers.

Retires AP-1 and AD-1 in docs/architecture/retail-divergence-register.md:
both rows described production zero-delta placement routing remaining on
the legacy resolver pending the Slice 4B2/4B route cutover; that resolver
no longer exists, so the condition each row tracked is now structurally
false rather than merely narrowed. AP-145 (routed through the prior commit)
and this commit's AP-1/AD-1 together bring the section counts to 101 AP / 47
AD active rows.

Builds on the AP-145 fix (previous commit) — this commit's staged tree was
independently rebuilt and its four suites independently rerun on top of
that commit before this commit was created, in addition to the combined
rebuild/rerun below.

Full-solution build: 0 errors (21 pre-existing warnings, all unrelated).
Suite results (combined tree): Core 4270/4271 passed (1 skip; the single
DatSoundCacheTests concurrent-decode-dedup failure is a known load-sensitive
race, confirmed passing standalone and unrelated to this change), Runtime
1176/1176, Headless 86/86, App 4132/4135 (3 skips).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

tests/AcDream.App.Tests/Input/PlayerMovementPlacementTransactionTests.cs

@@ -15,8 +15,47 @@ public sealed class PlayerMovementPlacementTransactionTests
     private const uint PriorCell = 0x0101_0101u;
     private const uint DestinationCell = 0x0101_0102u;
 
+    /// <summary>
+    /// C5a (2026-08-05) re-point: <c>PlayerMovementController.CommitPreparedPosition</c>
+    /// is deleted (zero production callers — production placement now arms
+    /// the login-entry leash via <c>ArmConstraintLeashAtCommittedPlacement</c>,
+    /// called from <c>RuntimeLocalPlayerPhysicsPublicationState.ArmFirstEntryConstraintLeash</c>
+    /// AFTER <c>RuntimeSetPositionState.TryApplyDormantLocalActivationFinalCommit</c>
+    /// has already published the shared current cell — see that method's
+    /// doc comment). Architecture-review correction (A5, 2026-08-05): the two
+    /// deleted invariants did NOT relocate symmetrically.
+    /// <b>Render-root publish DID move</b> — it happens inside
+    /// <c>RuntimeSetPositionState.TryApplyDormantLocalActivationFinalCommit</c>
+    /// itself, via <c>_physics.Engine.UpdatePlayerCurrCell(result.CellId)</c>
+    /// (<c>RuntimeSetPositionState.cs:2774</c>).
+    /// <b>Sticky-target release did NOT move anywhere</b> — <c>grep -rn
+    /// "UnStick" src/AcDream.Runtime/</c> returns zero call sites on this
+    /// path; the only local-player <c>UnStick</c> left is
+    /// <c>PlayerMovementController.SetPositionCore</c>'s
+    /// <c>if (publishSharedState) PositionManager?.UnStick();</c>, and
+    /// <c>PreparePositionForCommit</c> passes <c>publishSharedState: false</c>.
+    /// The unstick-at-first-entry-commit behaviour <c>CommitPreparedPosition</c>
+    /// used to perform is simply gone — it was dead code (the method had zero
+    /// production callers before this slice deleted it), so nothing regresses
+    /// today, but no layer pins the invariant "first-entry commit releases any
+    /// sticky target" any more. That is disposition 3.6's one real coverage
+    /// loss (see the C5a commit message).
+    /// This App-layer fixture (a bare <see cref="PlayerMovementController"/> +
+    /// <see cref="EntityPhysicsHost"/> pair, no Runtime activation pipeline)
+    /// cannot drive the Runtime final-commit transaction without
+    /// disproportionate fixture growth — recorded here as the explicit C5a
+    /// scope decision rather than silently dropped.
+    /// What THIS test still proves, unchanged: <see cref="PlayerMovementController.PreparePositionForCommit"/>
+    /// defers the render-root publish exactly as before (<c>controller.CellId</c>
+    /// updates immediately; <c>physics.DataCache.CellGraph.CurrCell</c> and
+    /// the sticky target stay untouched), and the surviving production
+    /// commit-arm, <see cref="PlayerMovementController.ArmConstraintLeashAtCommittedPlacement"/>,
+    /// arms the constraint leash without independently touching either —
+    /// neither the render-root publish (a DIFFERENT layer's job now) nor the
+    /// sticky release (nobody's job any more).
+    /// </summary>
     [Fact]
-    public void PreparedPosition_DoesNotPublishRenderRootOrUnstickUntilCommit()
+    public void PreparedPosition_DefersRenderRootPublish_CommitArmsLeashOnly()
     {
         PhysicsEngine physics = PhysicsWithCells(PriorCell, DestinationCell);
         physics.UpdatePlayerCurrCell(PriorCell);
@@ -38,11 +77,19 @@ public sealed class PlayerMovementPlacementTransactionTests
         Assert.Equal(DestinationCell, controller.CellId);
         Assert.Equal(PriorCell, physics.DataCache!.CellGraph.CurrCell!.Id);
         Assert.Equal(TargetGuid, player.PositionManager.GetStickyObjectId());
+        Assert.False(player.PositionManager.IsFullyConstrained());
+        Assert.Null(player.PositionManager.Constraint);
 
-        controller.CommitPreparedPosition();
+        controller.ArmConstraintLeashAtCommittedPlacement();
 
-        Assert.Equal(DestinationCell, physics.DataCache.CellGraph.CurrCell!.Id);
-        Assert.Equal(0u, player.PositionManager.GetStickyObjectId());
+        // The leash is armed...
+        Assert.NotNull(player.PositionManager.Constraint);
+        Assert.True(player.PositionManager.Constraint!.IsConstrained);
+        // ...but the surviving App-layer commit does NOT publish the render
+        // root or release the sticky target — those now happen inside the
+        // Runtime dormant-activation final commit, before this method runs.
+        Assert.Equal(PriorCell, physics.DataCache.CellGraph.CurrCell!.Id);
+        Assert.Equal(TargetGuid, player.PositionManager.GetStickyObjectId());
     }
 
     private static PhysicsEngine PhysicsWithCells(params uint[] cellIds)