feat(client): follow-facing — body commits to path, mouse drives layered head/torso look-at (T-1088)

During an RMB path-follow the body yaw locks to the leg direction (rig
commit_body_to_motion, from interpolated velocity — immune to SetFacing
interleaving between throttled steps). The mouse instead drives a layered
look-at: LookAtModifier3D pair on Head (±70°) + spine_02 (±30° torso twist
for looking far lateral/behind — past their sum the character physically
cannot look further without turning). Forward axis measured from the
armature rest pose (+Z), not guessed. Influence fades in/out on follow
start/end. Pure client presentation per D-249; SetFacing still rides the
wire, so the server vision cone follows the mouse while walking — the
character looks where the player points.

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

client/scripts/sandbox/locomotion_sandbox.gd

@@ -41,6 +41,11 @@ var _aim_provider: SandboxMouseAimProvider = null
 ## per frame). Installed in _ready(); the RMB gesture handling is in _unhandled_input.
 var _path_preview: PathPreview = null
 var _path_follower: PathFollower = null
+## T-1088 follow-facing: layered head/torso look-at (active during follows) and
+## its shared aim-target node (WorldRoot-local, positioned at the mouse ground
+## point each frame).
+var _head_look: SandboxHeadLook = null
+var _head_aim_target: Node3D = null
 ## Time (ms) a non-double RMB press began, -1 = none pending. Resolves click vs
 ## long-press on RELEASE (RMB_LONG_PRESS_MS threshold).
 var _rmb_press_msec: int = -1
@@ -97,6 +102,15 @@ func _ready() -> void:
 	_anim = LocomotionAnim.new()
 	_anim.setup(player_rig, character_visual)
 
+	# (3d) Follow-facing look-at (T-1088): aim target under WorldRoot (its local
+	# space IS the sim-aligned space ground_hit_local returns) + the modifier
+	# pair on the freshly-built skeleton.
+	_head_aim_target = Node3D.new()
+	_head_aim_target.name = "HeadAimTarget"
+	world_root.add_child(_head_aim_target)
+	_head_look = SandboxHeadLook.new()
+	_head_look.setup(character_visual.get_skeleton(), _head_aim_target)
+
 	# (4) InputMapper facing provider (design §9).
 	_install_facing_provider()
 
@@ -390,6 +404,32 @@ func _per_frame_update(delta: float) -> void:
 		_path_preview.update(char_tile, is_floor, greybox.store.size())
 		_path_follower.tick(char_tile, is_floor)
 
+	# T-1088 follow-facing: while a follow is active the body commits to the leg
+	# direction and the mouse drives the layered head/torso look-at instead.
+	# SetFacing still rides the wire (the server vision cone follows the mouse).
+	var follow_active := _path_follower.is_active()
+	player_rig.commit_body_to_motion = follow_active
+	_head_look.set_active(follow_active and not _input_suppressed())
+	_update_head_aim_target()
+	_head_look.update(delta)
+
+
+## Position the shared look-at target at the mouse's ground point (WorldRoot-
+## local — the same sim-aligned space the aim provider returns) at eye height.
+func _update_head_aim_target() -> void:
+	var cam := get_viewport().get_camera_3d()
+	if cam == null or _head_aim_target == null:
+		return
+	var mouse := get_viewport().get_mouse_position()
+	var hit: Variant = SandboxMouseAimProvider.ground_hit_local(
+		cam.project_ray_origin(mouse), cam.project_ray_normal(mouse),
+		world_root.global_transform
+	)
+	if hit != null:
+		var p: Vector3 = hit
+		p.y = SandboxConstants.HEAD_LOOK_EYE_HEIGHT_M
+		_head_aim_target.position = p
+
 
 ## Hard camera-pivot snap (design §7) — first-snapshot latch + teleport fan-out.
 func _snap_camera_to_player() -> void: