Initial K3 snapshot: 0.5B KDA/MLA/MoE train path

Standalone tree split from LLMRL/projects/kda. Includes Triton dt_bias
backward fix, train_k3 --preset 0.5b, SFT, Docker runtime, and tests.

tests/integration/test_train_overfit.py

@@ -0,0 +1,32 @@
+"""L7: toy overfit smoke test. 320 steps loss < 0.1."""
+import torch
+
+from kda.models.causal_lm import CausalLM
+from kda.models.config import KDAConfig
+
+
+def test_overfit():
+    cfg = KDAConfig()                                # 起步默认 toy 配置
+    torch.manual_seed(30)
+    model = CausalLM(cfg).cuda()
+
+    x = torch.randint(0, cfg.vocab_size, (4, 16), device="cuda")
+    labels = x.clone()
+    # A single repeated batch is an optimizer/dataflow smoke test, so converge it quickly.
+    optim = torch.optim.AdamW(model.parameters(), lr=1e-3, weight_decay=0.01)
+
+    for step in range(320):
+        optim.zero_grad()
+        loss = model(x, labels=labels)
+        loss.backward()
+        optim.step()
+        if step % 64 == 0 or step == 319:
+            print(f"  step {step:3d}  loss {loss.item():.4f}")
+
+    final = loss.item()
+    assert final < 0.1, f"final loss {final:.4f} > 0.1"
+    print(f"L7 overfit: PASSED (final loss {final:.4f})")
+
+
+if __name__ == "__main__":
+    test_overfit()