fix(ci): patch gpu_backend by dotted path in the live-detection routing test

`HardwareProfile.detect()` dereferences `routing._gb.has_cuda_rust()`, where
`_gb` is the `qector_decoder_v3.gpu_backend` entry in sys.modules. The test
patched the package-level attribute `qd.gpu_backend` instead. Locally the two
are the same object so it worked; in CI the patch did not take and `detect()`
kept reading the real hardware, so the two cases expecting gpu=True/cuda=True
failed on the GPU-less runner while passing on any developer machine with a GPU.

monkeypatch's dotted-path form resolves the module through sys.modules, which is
exactly what the code under test dereferences.

Author: qectorlab

python/tests/test_routing.py

@@ -137,8 +137,14 @@ def test_code_object_is_classified_structurally():
 # ---------------------------------------------------------------------------
 @pytest.mark.parametrize("cuda_rust,gpu", [(False, False), (False, True), (True, True)])
 def test_detected_hardware_drives_huge_batch(monkeypatch, cuda_rust, gpu):
-    monkeypatch.setattr(qd.gpu_backend, "has_cuda_rust", lambda: cuda_rust)
-    monkeypatch.setattr(qd.gpu_backend, "gpu_available", lambda: gpu)
+    # Patch the canonical module by dotted path, NOT via `qd.gpu_backend`.
+    # `HardwareProfile.detect()` reads `routing._gb.has_cuda_rust()`, where `_gb`
+    # is `qector_decoder_v3.gpu_backend` bound from sys.modules. The package-level
+    # attribute `qd.gpu_backend` is not reliably that same object - in CI it is
+    # shadowed, so patching it left `detect()` reading the real hardware and these
+    # two cases failed on any machine without a GPU while passing on one with.
+    monkeypatch.setattr("qector_decoder_v3.gpu_backend.has_cuda_rust", lambda: cuda_rust)
+    monkeypatch.setattr("qector_decoder_v3.gpu_backend.gpu_available", lambda: gpu)
 
     prof = HardwareProfile.detect()
     assert prof.cuda_rust is cuda_rust and prof.gpu is gpu