build(SM70): precompile GDN CUDA kernel to .so during docker build

precompile_gdn.py: calls torch.utils.cpp_extension.load with build_directory
to produce .so at build time. If build env has no GPU/compiler, fails
gracefully — kernel JIT compiles at runtime instead.

fused_fwd.py: _load_ext() now checks build/ dir for precompiled .so first,
skips 2-minute JIT compilation if found.

qwen3_6_scripts/flash_qla_sm70/fused_fwd.py

@@ -20,6 +20,24 @@ def _load_ext():
         raise RuntimeError("SM70 FlashQLA backend requires CUDA.")
 
     os.environ.setdefault("TORCH_CUDA_ARCH_LIST", "7.0;7.5")
+
+    # Try precompiled .so first (built during docker build)
+    build_dir = Path(__file__).with_name("build")
+    if build_dir.is_dir():
+        so_files = list(build_dir.glob("*.so"))
+        if so_files:
+            try:
+                _EXT = load(
+                    name="flash_qla_sm70_gdn_strided",
+                    sources=[],  # empty — just load from build_directory
+                    build_directory=str(build_dir),
+                    verbose=False,
+                )
+                return _EXT
+            except Exception:
+                pass  # fall through to JIT
+
+    # JIT compile (slow, ~2min first time)
     src = Path(__file__).with_name("csrc") / "gdn_forward.cu"
     _EXT = load(
         name="flash_qla_sm70_gdn_strided",