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/tests/test_weights.nim

@@ -4,6 +4,7 @@
 import std/[os, math]
 import arraymancer
 import "../network"
+import "../training"
 import "../weights"
 
 template check(cond: bool, msg: string) =
@@ -67,18 +68,21 @@ block testLoadBest:
 
   # Try with no weights — should return false
   var acEmpty = initActorCritic()
-  check not loadBestAvailable(acEmpty, root), "should return false with no weights"
+  var adamEmpty: ACAdamStates
+  check not loadBestAvailable(acEmpty, adamEmpty, root).loaded, "should return false with no weights"
 
   # Save to latest/; should load
   saveWeights(ac0, root / "latest")
   var ac1 = initActorCritic()
-  check loadBestAvailable(ac1, root), "should load from latest/"
+  var adam1: ACAdamStates
+  check loadBestAvailable(ac1, adam1, root).loaded, "should load from latest/"
 
   # Remove latest/, save to checkpoint_1/ — should fall back
   removeDir(root / "latest")
   saveWeights(ac0, root / "checkpoint_1")
   var ac2 = initActorCritic()
-  check loadBestAvailable(ac2, root), "should load from checkpoint_1/"
+  var adam2: ACAdamStates
+  check loadBestAvailable(ac2, adam2, root).loaded, "should load from checkpoint_1/"
 
   removeDir(root)