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.

kda/_fla/ops/backends/__init__.py

@@ -0,0 +1,34 @@
+"""Identity dispatch: keep the NVIDIA Triton implementation in this tree."""
+
+from __future__ import annotations
+
+from collections.abc import Callable
+from typing import TypeVar
+
+F = TypeVar("F", bound=Callable)
+
+
+def dispatch(operation: str):
+    def decorator(func: F) -> F:
+        return func
+
+    return decorator
+
+
+class BaseBackend:
+    backend_type = "triton"
+
+    def is_available(self) -> bool:
+        return True
+
+    def is_enabled(self) -> bool:
+        return True
+
+
+class BackendRegistry:
+    @classmethod
+    def ensure_initialized(cls, operation: str) -> None:
+        return None
+
+
+__all__ = ["BackendRegistry", "BaseBackend", "dispatch"]