feat(*): Replace the test terrain arena with an enclosed standard-size space-platform arena (shared ArenaBoundary floor/walls/ceiling scene, starfield sky, ball CCD) and derive TrainingMode placement bounds from it, dropping the out-of-bounds reward guard

Game/scripts/arena_boundary.gd

@@ -0,0 +1,13 @@
+class_name ArenaBoundary
+extends StaticBody3D
+
+# The standard arena play volume (inner faces of the enclosure). Every arena
+# instances objects/arena_boundary.tscn so all arenas share one size; code
+# that needs field dimensions derives them from these constants rather than
+# restating numbers.
+const INNER_HALF_X := 12.0
+const INNER_HALF_Z := 18.0
+const INNER_HEIGHT := 12.0
+# Goal-centre distance from arena centre; the end walls sit 1 m behind, so a
+# ball pinned against them still overlaps the goal sensor.
+const GOAL_LINE_Z := 17.0