fix #180: port retail's stateful camera sought-position - the sweep target converges onto the wall contact

The camera-collision sweep strobed the eye 0.27m every ~5-10 frames while
the compressed chase boom moved along corridor walls (pulledIn 0.27<->0.53
on ~1.4mm input drift): RetailChaseCamera re-demanded the FULL-length ideal
boom from scratch each frame, so the pivot->eye ray re-rolled the same
knife-edge r+-eps graze on the double-faced slabs every frame, and its two
first-contact solutions tear-interleaved at ~1700fps into the #176
"stripes/triangles".

Retail never re-rolls that ray. CameraManager::UpdateCamera (0x00456660)
interpolates FROM THE CURRENT SWEPT VIEWER toward the desired pose
(interpolate_origin/rotation, stiffness 0.45 x dt x 10, clamped) and the
result becomes viewer_sought_position (SmartBox::PlayerPhysicsUpdatedCallback
0x00452d60); update_viewer (0x00453ce0) sweeps pivot->SOUGHT. Pressed against
a wall the sweep ray extends one interpolation step past the contact
(sub-mm at high fps), so a bistable graze can move the eye by at most that
step - the strobe is structurally impossible. A 0.4mm/2e-4 dead-band parks
the sought exactly on the viewer when converged (0x00456fcd-0x00457035).

- RetailChaseCamera: _dampedEye -> _soughtEye + _publishedEye (retail's two
  Positions); lerp base = the published (swept) viewer; sweep targets the
  sought; total-fallback (ViewerCellId==0) resets the sought like
  set_viewer(player_pos, 1). The old "collision must NOT feed back into the
  damped state" comment had the coupling backwards - what stays clean is the
  transient desired pose, not the sought.
- SweepEye untouched (faithful update_viewer port, exonerated by the #180
  investigation).
- Tests: the old pin asserting instant full re-extension after a clamp
  (the divergence itself) replaced with four retail pins: gradual
  re-extension, sweep-target-converges-onto-contact, total-fallback
  re-extends from the player, wall-press glide stability.
- Pseudocode doc: docs/research/2026-07-06-camera-sought-position-pseudocode.md
  (UpdateCamera tail incl. the sought derivation + set_viewer reset semantics
  + Frame interpolate/close_rotation).
- Register: AD-37 (forward-vector nlerp vs quaternion slerp), AD-38
  (init-at-full-extension vs retail re-extend-from-player) - both
  pre-existing, identified during the decomp reading.

Suites green (Core 2599+2skip / App 729+2skip / UI 425 / Net 385).
Pending: autonomous visual verify + user gate (#180 + the #176 re-gate).

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

docs/architecture/retail-divergence-register.md

@@ -96,6 +96,8 @@ accepted-divergence entries (#96, #49, #50).
 | AD-34 | Retail's intrusive `DLListBase`/`DLListData` lists are managed `LinkedList<T>`s; node identity via `LinkedListNode<>` references (Phase R1 anim list, 2026-07-02; extended R2-Q3 to `MotionTableManager.pending_animations`; extended R4-V2 to `MoveToManager.pending_actions` — whose node type, retail `MoveToManager::MovementNode`, is RENAMED `MoveToNode` to avoid colliding with R2's `Motion/MotionNode.cs` pending_motions node) | `src/AcDream.Core/Physics/Motion/CSequence.cs` (`_animList`); `src/AcDream.Core/Physics/Motion/MotionTableManager.cs` (`_pendingAnimations`); `src/AcDream.Core/Physics/Motion/MoveToManager.cs` (`_pendingActions`) + `MoveToNode.cs` | Same topology + cursor semantics (curr_anim/first_cyclic/tail-anchored scans are node references); unlink/delete becomes `Remove(node)`; conformance tests pin the surgery state tables | Any retail behavior depending on the −4 pointer adjustment or node memory reuse (none observed in the decomp) would diverge; a reader grepping for retail's `MovementNode` name must find it via this row | `acclient.h` DLListBase; `r1-csequence-decomp.md` §0; `r2-motiontable-decomp.md` §11; `r4-moveto-decomp.md` node factories §4a |
 | AD-35 | `MotionTableManager.PerformMovement`'s unhandled-type default case returns the named sentinel `0xFFFFFFFF` (`MotionTableManagerError.NotHandled`); retail's compiled code leaks the `CSequence*` pointer reinterpreted as the return code (BN-confirmed artifact, dead/unreachable — callers gate on type first) (R2-Q3, 2026-07-02) | `src/AcDream.Core/Physics/Motion/MotionTableManager.cs` (`PerformMovement` default case) | No retail caller consults the return value for unhandled types (RawCommand/StopRawCommand/MoveTo\*/TurnTo\* route elsewhere); returning a stable non-zero sentinel preserves the only observable contract (non-zero = not success) without fabricating a pointer-shaped number | If a future port wires a caller that passes unhandled types AND branches on the exact return value, it would see `0xFFFFFFFF` where retail saw an arbitrary pointer — flag at that port | `PerformMovement` 0x0051c0b0 (`r2-motiontable-decomp.md` §11 default-case note) |
 | AD-36 | `IMotionDoneSink.MotionDone` consumed for CREATURE-class entities only: R3-W2 binds the seam to the entity's `MotionInterpreter.MotionDone` (player via `PlayerMovementController.Motion`, remotes via `RemoteMotion.Motion`, resolved at fire time); interp-less entities (statics that never receive a UM/UP and so never get a `RemoteMotion`) keep a diagnostic-recorder-only target — retail gives every CPhysicsObj a MovementManager/CMotionInterp (R2-Q4 seam, narrowed R3-W2, 2026-07-02; R5-V5 gave every `RemoteMotion`/player ONE literal `MovementManager` facade, so the residue is only the no-RemoteMotion class) | `src/AcDream.App/Rendering/GameWindow.cs` (TickAnimations MotionDoneTarget bind) | Motion for entities without a `RemoteMotion` (never UM/UP-touched) completes via the manager queue alone; nothing consumes their MotionDone until every sequencer-owning entity gets a host/`RemoteMotion` (doors DO have one since the R4-V5 door fix — first UM creates it) | An entity behavior depending on pending_motions bookkeeping in that no-RemoteMotion class (none known) would silently no-op | `CPhysicsObj::MotionDone` 0x0050fdb0; retire when every sequencer-owning entity constructs a `RemoteMotion`/host (post-M1.5 entity-class unification; R5-V5 closed the facade half) |
+| AD-37 | Camera rotation state is a forward VECTOR (nlerp + normalize; roll always 0, up = world Z); retail's sought carries a full Frame and slerps quaternions (`Frame::interpolate_rotation` shortest-path slerp with 2e-4 nlerp fallback). The dead-band compares forward-vector distance against the same 2e-4 epsilon retail applies per quaternion component | `src/AcDream.App/Rendering/RetailChaseCamera.cs` (`_dampedForward`, `ApplyConvergenceSnap`) | The chase camera never rolls (heading frames are Z-up by construction), so a forward vector spans the reachable rotation space; identified (not introduced) during the #180 UpdateCamera tail reading | If a future camera mode needs roll (death cam, cutscene) the vector state can't represent it; large-angle per-frame turns nlerp (chord) vs slerp (arc) — imperceptible at 0.45-stiffness step sizes | `Frame::interpolate_rotation` 0x00535390, `Frame::close_rotation` 0x00455d70; pseudocode doc 2026-07-06-camera-sought-position |
+| AD-38 | First camera frame seats sought = viewer = the full-length target eye (starts converged); retail hard resets both to the PLAYER's position (`set_viewer(pos, reset_sought=1)`) at login/teleport/cell-loss, so the retail camera visibly re-extends outward from the head over ~1 s | `src/AcDream.App/Rendering/RetailChaseCamera.cs` (`_initialised` branch) | acdream teleports don't reset the camera object today; the pivot-anchored sweep self-heals in one convergence, and the login zoom-out is cosmetic. Identified (not introduced) during the #180 reading | Login/teleport first frames show a fully-extended boom where retail shows a zoom-out; if a teleport lands the stale sought behind distant geometry the first sweeps clip until convergence (sub-second) | `SmartBox::set_viewer` 0x00452c40 (reset_sought=1 sites pc:92775, pc:92886); retire by wiring teleport → sought reset |
 
 ---