feat(arena): add elevated-goal arena variants

Each of the three arenas gains an ELEVATED sibling scene that inherits the
base arena and overrides the boundary's goal_mode, the two goal transforms
and the ship spawns, so the goal sits at mid-wall height instead of flush
with the deck. Registered in ArenaRegistry alongside the floor-level arenas,
plus a training_elevated scene for self-play on them.

TrainingMode reads the arena's goal_mode once in _start() and widens the
ball-placement height range to match the goal's real position; on FLOOR
arenas the bound is a no-op, so floor-level training is unchanged. It is read
in _start() rather than _ready() because TrainingMode has no _ready()
override and GameMode._ready() is what discovers the arena first.

Policies trained against floor-level goals are not expected to score on an
elevated one, so the two are kept as separate arenas rather than a variant of
the same entry.

Game/scripts/arena_registry.gd

@@ -1,14 +1,26 @@
 class_name ArenaRegistry
 
 # Single source of truth for available arenas: the Free Play menu dropdown
-# lists these, and Match/Spectate pick one at random each session. Training
-# is exempt — training.tscn keeps its own fixed arena_01 child.
+# lists all of these, and Match/Spectate pick at random each session from
+# those with "random" true. Training is exempt — training.tscn/
+# training_elevated.tscn each keep their own fixed Arena child.
+#
+# Elevated-goal variants are Free-Play-only ("random": false) until a
+# checkpoint trained on training_elevated.tscn is promoted — the current
+# promoted bots (see main_menu.gd's DIFFICULTIES) were trained exclusively
+# on floor-level goals and can't be expected to score on an elevated one.
+# Flip a variant's "random" flag to true once that training/promotion has
+# happened.
 const ARENAS := [
-	{"name": "Starfield", "path": "res://scenes/arena_01.tscn"},
-	{"name": "Nebula", "path": "res://scenes/arena_02.tscn"},
-	{"name": "Asteroid Field", "path": "res://scenes/arena_03.tscn"},
+	{"name": "Starfield (Floor Goals)", "path": "res://scenes/arena_01.tscn", "random": true},
+	{"name": "Starfield (Elevated Goals)", "path": "res://scenes/arena_01_elevated.tscn", "random": false},
+	{"name": "Nebula (Floor Goals)", "path": "res://scenes/arena_02.tscn", "random": true},
+	{"name": "Nebula (Elevated Goals)", "path": "res://scenes/arena_02_elevated.tscn", "random": false},
+	{"name": "Asteroid Field (Floor Goals)", "path": "res://scenes/arena_03.tscn", "random": true},
+	{"name": "Asteroid Field (Elevated Goals)", "path": "res://scenes/arena_03_elevated.tscn", "random": false},
 ]
 
 
 static func random_path() -> String:
-	return ARENAS[randi() % ARENAS.size()]["path"]
+	var candidates := ARENAS.filter(func(arena): return arena["random"])
+	return candidates[randi() % candidates.size()]["path"]

Game/scripts/training_mode.gd

@@ -120,8 +120,16 @@ var _ai_overrides := {}
 # arena spawn instead of drifting into the play area as a stray obstacle.
 var _inert_ships: Array[Ship] = []
 
+# Whether this run's arena is an ELEVATED-goal variant (see
+# ArenaBoundary.GoalMode) — read once _start() runs so _place_ball_near_goal
+# can sample a height range matching the goal's real position. Read here
+# rather than _ready(): TrainingMode has no _ready() override, and
+# GameMode._ready() is what discovers `arena` before calling _start().
+var _elevated := false
+
 
 func _start() -> void:
+	_elevated = (arena.get_node("Boundary") as ArenaBoundary).goal_mode == ArenaBoundary.GoalMode.ELEVATED
 	_parse_eval_args()
 	_parse_curriculum_args()
 	spawn_ball()
@@ -371,9 +379,12 @@ func _place_ball_random() -> void:
 func _place_ball_near_goal() -> void:
 	var goal := _goal_for_team(0) if randf() < attack_goal_bias else _goal_for_team(1)
 	var toward_centre := -signf(goal.global_position.z)
+	# Upper Y bound is a no-op on FLOOR arenas (goal.global_position.y ~0.79,
+	# so maxf(4.0, ...) stays 4.0); on ELEVATED arenas it widens to sample
+	# near the goal's real height instead of always landing near the floor.
 	var position := Vector3(
 		randf_range(-4.0, 4.0),
-		randf_range(FIELD_MIN_Y, 4.0),
+		randf_range(FIELD_MIN_Y, maxf(4.0, goal.global_position.y + 2.0)),
 		goal.global_position.z + toward_centre * randf_range(3.0, 6.0)
 	)
 	var to_goal := (goal.global_position - position).normalized()