feat(SAC_LSTM_Bot): training harness (#49)

sac_train.sh orchestrates chunked self-play via tools/training_runner/
RunTraining.java: weighted opponent sampling per chunk, deterministic
eval (SACLSTM_EVAL_MODE=1) every N chunks with win-rate tracking, best
checkpoint (weights/sac_best.zip) by eval score, crash-restart loop on
the runner's liveness detection.

Supporting changes:
- integration.nim: opponentKey() keys the NewBattle buffer-clear rule on
  getBotName(id) with numeric-id fallback (#49 Q14 follow-up);
  bumpRoundCounter() emits the per-round liveness signal.
- SAC_LSTM_Bot.nim: onRoundEnded -> bumpRoundCounter().
- RunTraining.java: BOT_NAME env parameterizes result matching
  (default PPO_Bot, unchanged behavior for PPO).
- Launch packaging: root SAC_LSTM_Bot.json + .sh for the booter;
  src json name aligned to 'SAC_LSTM_Bot' so self-reported identity
  matches the booted identity (mismatch = runner connect timeout).

SAC_LSTM_Bot/tests/test_integration.nim

@@ -3,7 +3,8 @@
 ## drain-then-train NewBattle/Shutdown handling, trainPass safety below canSample.
 
 import arraymancer except Linear
-import std/locks
+import std/[locks, json, os]
+import tankroyale_botapi   # updateBotNames: seed the vendored id->name table
 import SAC_LSTM_Bot/integration
 import SAC_LSTM_Bot/state    # STATE_DIM
 import SAC_LSTM_Bot/actions  # ACTION_DIM
@@ -46,14 +47,20 @@ block:
   assert not ok4, "closed channel must report dataAvailable=false"
   echo "PASS TrainingMsg channel round-trip"
 
-# ── 2. NewBattle clears only when the opponent changes; Shutdown stops ────────
+# ── 2. NewBattle clears only when the opponent NAME changes; Shutdown stops ───
 block:
+  # Seed the API's id->name table (v1.0.1) for name-based identity (#49).
+  updateBotNames(parseJson(
+    """{"bots":[{"id":7,"name":"Corners"},{"id":8,"name":"Crazy"}]}"""))
+  assert opponentKey(7) == "Corners", "known id resolves to name"
+  assert opponentKey(99) == "99", "unknown id falls back to numeric string"
+
   var st: TrainState
   st.trainer = initSACTrainer(STATE_DIM, ACTION_DIM)
   st.buf = newReplayBuffer(64, STATE_DIM, ACTION_DIM, burnIn = 2, trainWindow = 3)
 
   assert handleTrainingMsg(st, TrainingMsg(kind: tmkNewBattle, enemyId: 7))
-  assert st.lastEnemyId == 7
+  assert st.lastEnemyKey == "Corners"
 
   for i in 0 ..< 5:
     var m = TrainingMsg(kind: tmkTransition)
@@ -68,11 +75,24 @@ block:
   # Opponent changed -> clear.
   assert handleTrainingMsg(st, TrainingMsg(kind: tmkNewBattle, enemyId: 8))
   assert st.buf.len == 0, "opponent change must clear"
-  assert st.lastEnemyId == 8
+  assert st.lastEnemyKey == "Crazy"
+
+  # Nameless window (pre-BotListUpdate) is a distinct key.
+  assert handleTrainingMsg(st, TrainingMsg(kind: tmkNewBattle, enemyId: 99))
+  assert st.lastEnemyKey == "99" and st.buf.len == 0
 
   # Shutdown stops the caller's loop.
   assert not handleTrainingMsg(st, TrainingMsg(kind: tmkShutdown))
-  echo "PASS drain-then-train NewBattle/Shutdown"
+  echo "PASS name-keyed NewBattle clear / Shutdown"
+
+# ── 2b. bumpRoundCounter increments per call, cold-starts at 1 ────────────────
+block:
+  let tmp = getTempDir() / "sac_test_weights_" & $getCurrentProcessId()
+  putEnv("SACLSTM_WEIGHTS_PATH", tmp / "sac_latest.zip")
+  bumpRoundCounter()
+  bumpRoundCounter()
+  assert readFile(tmp / "round_counter.txt") == "2", "counter increments per round"
+  echo "PASS bumpRoundCounter"
 
 # ── 3. trainPass is a safe no-op below canSample (no steps, no publish) ───────
 block: