Add CUDA 12.9 builds to test channel and aarch64 matrix

Add CUDA 12.9 to the test channel in CUDA_ARCHES_DICT (for validating
release candidate binaries) and add 12.9-aarch64 to CUDA_AARCH64_ARCHES.
Also adds the corresponding CUDA_CUDNN_VERSIONS entry (12.9.1 / cudnn 9),
matching how 12.9 was previously configured.

Reference test assets are regenerated via
'python -m tools.tests.test_generate_binary_build_matrix --update-reference-files'.
This regeneration also picks up a pre-existing staleness on this branch:
the assets had stable_version 2.11.0 while the script already declares
CURRENT_STABLE_VERSION = 2.12.0, which had left the 7 matrix tests red.
The regenerated assets now pass.

Author: atalman

tools/scripts/generate_binary_build_matrix.py

@@ -36,7 +36,7 @@
 }
 CUDA_ARCHES_DICT = {
     "nightly": ["12.6", "13.0", "13.2"],
-    "test": ["12.6", "13.0", "13.2"],
+    "test": ["12.6", "12.9", "13.0", "13.2"],
     "release": ["12.6", "12.8", "13.0"],
 }
 
@@ -49,6 +49,7 @@
 CUDA_CUDNN_VERSIONS = {
     "12.6": {"cuda": "12.6.3", "cudnn": "9"},
     "12.8": {"cuda": "12.8.0", "cudnn": "9"},
+    "12.9": {"cuda": "12.9.1", "cudnn": "9"},
     "13.0": {"cuda": "13.0.0", "cudnn": "9"},
     "13.2": {"cuda": "13.2.0", "cudnn": "9"},
 }
@@ -59,7 +60,7 @@
     "release": "13.0",
 }
 
-CUDA_AARCH64_ARCHES = ["12.6-aarch64", "13.0-aarch64", "13.2-aarch64"]
+CUDA_AARCH64_ARCHES = ["12.6-aarch64", "12.9-aarch64", "13.0-aarch64", "13.2-aarch64"]
 
 PACKAGE_TYPES = ["wheel", "libtorch"]
 CXX11_ABI = "cxx11-abi"