chore: rename libs→common_libs, all bot dirs to _garage suffix, fix all path refs

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

PPO_Bot_garage/tests/diag_ppo_fullbuffer.nim

@@ -0,0 +1,36 @@
+## diag_ppo_fullbuffer.nim — run ppoUpdate on a FULL 8192-transition buffer.
+## Verifies whether the round-10 death (first ppoUpdate at buffer capacity)
+## is a real crash in ppoUpdate or purely the saveAdamStates empty-shape bug.
+
+import std/[math, random]
+import arraymancer
+import "../network"
+import "../training"
+
+var ac = initActorCritic()
+var adam: ACAdamStates  # uninitialised → ppoUpdate must reinit (fresh-process path)
+var buf = initTrajectoryBuffer()
+
+randomize(1)
+while buf.len < MAX_TRANSITIONS:
+  var t: Transition
+  for i in 0..<STATE_DIM: t.state[i] = rand(1.0'f32) - 0.5'f32
+  for i in 0..<ACTION_DIM: t.action[i] = rand(1.0'f32) - 0.5'f32
+  t.logProb = rand(1.0'f32) - 1.0'f32
+  t.reward = rand(0.02'f32) - 0.01'f32
+  t.value = rand(0.1'f32)
+  t.done = buf.len mod 300 == 299
+  buf.add(t)
+
+echo "buffer.len = ", buf.len, " (MAX=", MAX_TRANSITIONS, ")"
+var ac2 = ac
+var adam2: ACAdamStates
+try:
+  let m = ppoUpdate(ac2, buf, lastValue = 0.0'f32, adamStates = adam2)
+  echo "ppoUpdate OK: aLoss=", m.actorLoss, " vLoss=", m.valueLoss, " gNorm=", m.gradNorm
+except CatchableError as e:
+  echo "CAUGHT CatchableError: ", e.msg
+  echo getStackTrace(e)
+except Defect as e:
+  echo "CAUGHT Defect: ", e.msg
+  echo getStackTrace(e)