Fix property-pool rollback errors in headless server

The dedicated server was spawning a host avatar (peer 1) via
player-spawner._handle_host() and _handle_new_peer(1) on startup,
which registered RollbackSynchronizer subjects with the netfox
rollback system. Since no client input exists for peer 1, the
rollback system tried to record state for freed/recycled objects,
flooding the log with property-pool.gd errors.

Fix:
- Add spawn_host_avatar @export flag to player-spawner.gd
- Set it false in headless_server.gd before starting the ENet server
- PlayerSpawner._handle_host() skips avatar spawn when flag is false
- PlayerSpawner._handle_new_peer(1) skips avatar spawn when flag is false

Server startup is now clean: no property-pool errors, no RPC path errors.

examples/multiplayer-fps/scripts/player-spawner.gd

@@ -2,6 +2,10 @@ extends Node
 
 @export var player_scene: PackedScene
 
+## Whether to spawn an avatar for the host (peer 1) when on_server_start fires.
+## Set to false for dedicated server operation.
+@export var spawn_host_avatar: bool = true
+
 var t_spawn_points: Array[Marker3D] = []
 var ct_spawn_points: Array[Marker3D] = []
 var team_manager: TeamManager
@@ -43,12 +47,20 @@ func _handle_connected(id: int):
 	_spawn(id)
 
 func _handle_host():
+	if not spawn_host_avatar:
+		print("[PlayerSpawner] Dedicated server mode — skipping host avatar")
+		if round_manager:
+			round_manager.start_match()
+		return
 	_spawn(1)
 	# Auto-start match
 	if round_manager:
 		round_manager.start_match()
 
 func _handle_new_peer(id: int):
+	if id == 1 and not spawn_host_avatar:
+		print("[PlayerSpawner] Skipping spawn for peer 1 (dedicated server)")
+		return
 	_spawn(id)
 	if multiplayer.is_server():
 		team_manager.sync_to_peer(id)