feat(SAC_LSTM_Bot): campaign v2 levers — loss metrics, eval-mode gate, eval rotation + MA gating (part 1)

Levers 3, 4, 1 of the #57 sign-off (execution order 3->4->1), tracked in #59.

- Lever 3 (#59): one JSONL line per trainPass in training_metrics.jsonl with
  exactly the scalars sacUpdate already exposes (SACMetrics: critic/actor/alpha
  losses + alpha, averaged per pass) plus epoch, buffer size (replay_buffer.len),
  cumulative steps and drained count. No trainer change needed.
- Lever 4 (#59): sendTrainingMsg drops all training input while SACLSTM_EVAL_MODE=1
  (existing #49 harness mechanism) — eval battles can neither pollute the replay
  buffer nor trigger gradient updates; one-time stderr notice at bot init.
- Lever 1 (#59): sac_train.sh evaluates every SAC_EVAL_OPPONENTS entry per cycle
  (results carry opponent name in eval_log.jsonl); best-gating now uses a
  composite = mean over opponents of the last-5-evals moving average per
  opponent. best_score.txt format change: float composite replaces the
  single-opponent integer win rate semantics (retired).
- Tests: metricsLine JSONL scalars + eval-mode suppression asserts.

Refs: #59, #57

SAC_LSTM_Bot/tests/test_integration.nim

@@ -122,3 +122,29 @@ block:
   assert alpha == t0.alpha()
   discard tc1
   echo "PASS flat snapshot pack/unpack round-trip"
+
+# ── 5. Lever 3 (#59): metricsLine emits exactly the exposed trainer scalars ───
+block:
+  let line = metricsLine(1787394115.123, 42, 500, 20, 20,
+    SACMetrics(criticLoss: 0.5'f32, actorLoss: -1.5'f32,
+               alphaLoss: 0.25'f32, alpha: 2.0'f32))
+  let j = parseJson(line)   # throws on malformed JSONL
+  assert j["steps"].getInt() == 42 and j["buffer_size"].getInt() == 500
+  assert j["drained"].getInt() == 20 and j["grad_steps"].getInt() == 20
+  assert abs(j["critic_loss"].getFloat() - 0.5) < 1e-3
+  assert abs(j["actor_loss"].getFloat() + 1.5) < 1e-3
+  assert abs(j["alpha_loss"].getFloat() - 0.25) < 1e-3
+  assert abs(j["alpha"].getFloat() - 2.0) < 1e-3
+  assert j["epoch"].getFloat() > 1e9
+  echo "PASS metricsLine JSONL scalars"
+
+# ── 6. Lever 4 (#59): SACLSTM_EVAL_MODE=1 suppresses training input ───────────
+block:
+  putEnv("SACLSTM_EVAL_MODE", "1")
+  # Gate fires before any channel traffic: false = dropped, nothing enqueued.
+  assert not sendTrainingMsg(TrainingMsg(kind: tmkTransition)),
+    "eval mode must drop transitions"
+  assert not sendTrainingMsg(TrainingMsg(kind: tmkNewBattle, enemyId: 7)),
+    "eval mode must drop NewBattle (no buffer clears from eval)"
+  delEnv("SACLSTM_EVAL_MODE")
+  echo "PASS eval-mode training-input suppression"