fix(*): apply the locomotion mask during in-game/eval inference, not just training

AIShipController (eval + real gameplay) ran the raw policy output unmasked
regardless of allow_vertical/allow_pitch_roll, while ShipAIController
(training) correctly discarded those axes for grounded curriculum stages.
A grounded-trained model's untrained vertical/pitch-roll output reached the
ship as noise during eval, understating it against models that were never
handicapped this way.

training/curriculum.py

@@ -60,6 +60,7 @@ STAGES = [
             "--attack-goal-bias", "1.0",
             "--no-allow-vertical", "--no-allow-pitch-roll",
         ],
+        "grounded": True,
     },
     {
         "name": "defend",
@@ -67,14 +68,17 @@ STAGES = [
             "--opponent-mode", "self_play",
             "--no-allow-vertical", "--no-allow-pitch-roll",
         ],
+        "grounded": True,
     },
     {
         "name": "no_draws",
         "flags": ["--draw-penalty", "5"],
+        "grounded": False,
     },
     {
         "name": "mechanics",
         "flags": [],
+        "grounded": False,
     },
 ]
 
@@ -145,9 +149,20 @@ def run_stage_attempt(stage_index: int, attempt: int, args) -> str:
     return exp
 
 
-def evaluate_attempt(experiment: str, reference: str, episodes: int) -> dict:
+def reference_grounded(stage_index: int) -> bool:
+    # rookie.json predates the locomotion mask entirely — always full 3D.
+    return False if stage_index == 0 else STAGES[stage_index - 1]["grounded"]
+
+
+def evaluate_attempt(experiment: str, reference: str, episodes: int, stage_index: int) -> dict:
     candidate = TRAINING_DIR.parent / "Game" / "bots" / f"{experiment}.json"
     cmd = [".venv/bin/python", "evaluate.py", str(candidate), reference, "--episodes", str(episodes)]
+    # Must match how each side was actually trained — see ai_ship_controller.gd's
+    # allow_vertical/allow_pitch_roll and evaluate.py's --grounded-a/-b.
+    if STAGES[stage_index]["grounded"]:
+        cmd.append("--grounded-a")
+    if reference_grounded(stage_index):
+        cmd.append("--grounded-b")
     print(" ".join(cmd))
     subprocess.run(cmd, cwd=TRAINING_DIR, check=True)
     history = json.loads(EVAL_HISTORY_PATH.read_text())
@@ -217,7 +232,7 @@ def main():
 
         experiment = run_stage_attempt(stage_index, attempt, args)
         reference = reference_bot(stage_index)
-        record = evaluate_attempt(experiment, reference, EVAL_EPISODES)
+        record = evaluate_attempt(experiment, reference, EVAL_EPISODES, stage_index)
         decision = decide(record)
 
         print(f"{experiment}: candidate {record['wins_a']}-{record['wins_b']} reference "