feat(PPO_Bot): persist Adam optimizer state and round counter across restarts (#35)

Save ACAdamStates (m/v tensors + t counters) as .npy files alongside
network weights in latest/ and checkpoint dirs; save round counter to
round_counter.txt. loadBestAvailable restores both on startup; fresh
start works unchanged when files are absent.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

PPO_Bot/training.nim

@@ -71,9 +71,9 @@ proc computeGAE*(rewards, values: seq[float32];
 # ── Manual Adam state ─────────────────────────────────────────────────────────
 
 type
-  AdamState = object
-    m, v: Tensor[float32]
-    t: int
+  AdamState* = object
+    m*, v*: Tensor[float32]
+    t*: int
 
 proc initAdamState(like: Tensor[float32]): AdamState =
   result.m = zeros_like(like)
@@ -130,9 +130,9 @@ proc mlpBackward(mlp: MLP; fwd: MLPFwd; x: Tensor[float32];
 
 type ACAdamStates* = object
   ## One AdamState per learnable tensor in ActorCritic.
-  aw1, ab1, aw2, ab2, aw3, ab3: AdamState  # actor MLP
-  cw1, cb1, cw2, cb2, cw3, cb3: AdamState  # critic MLP
-  logStd: AdamState
+  aw1*, ab1*, aw2*, ab2*, aw3*, ab3*: AdamState  # actor MLP
+  cw1*, cb1*, cw2*, cb2*, cw3*, cb3*: AdamState  # critic MLP
+  logStd*: AdamState
   initialized*: bool
 
 proc initACAdamStates*(ac: ActorCritic): ACAdamStates =