feat(PPO_Bot): multi-round transition accumulation (UPDATE_INTERVAL=10)

- Accumulate transitions across 10 rounds (~3000) before PPO update
  (was per-round ~300 — gradient estimates were far too noisy)
- training.nim: MAX_TRANSITIONS 4096→8192, done flag on transitions,
  GAE handles episode boundaries correctly
- PPO_Bot.nim: buffer persists across rounds, update every N rounds
- training.env: lr 5e-5→1e-4, entropy 0.001, UPDATE_INTERVAL=10

tools/training_runner/training.env

@@ -1,9 +1,10 @@
 # TRAINING_OPPONENT=Fire
 TRAINING_ROUNDS=30000
 PPOB_LOG_FILE=/home/davide/Projects/SirRoboGarage/tools/training_runner/logs/fire_training.jsonl
-PPOB_LR=0.0003
+PPOB_LR=1e-4
+PPOB_UPDATE_INTERVAL=10
 PPOB_CLIP_EPSILON=0.2
-PPOB_ENTROPY_COEFF=0.01
+PPOB_ENTROPY_COEFF=0.001
 PPOB_VALUE_LOSS_COEFF=0.5
 PPOB_MAX_GRAD_NORM=0.5
 PPOB_GAMMA=0.99
@@ -11,6 +12,7 @@ PPOB_LAM=0.95
 PPOB_EPOCHS=4
 PPOB_MINI_BATCH_SIZE=64
 PPOB_LOG_STD_FLOOR=-3.0
-PPOB_INITIAL_LOG_STD=0.0
+PPOB_LOG_STD_CEILING=0.0
+PPOB_INITIAL_LOG_STD=-0.5
 
 # PPOB_EVAL_ONLY=1 → freeze training (pure evaluation)