[build] Update CUDA arch for CUDA 13.x compatibility

CUDA 13.x dropped support for compute capability < 7.0.
Updated to support Volta (7.0) through Hopper (9.0).

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: Breakthrough

setup.py

@@ -23,6 +23,9 @@ def main():
     build_rolling = get_build_env_var_by_name("rolling")
 
     # TODO(@Breakthrough): What architectures should we ship?
+    # TODO(@Breakthrough): Create a separate builder/shard using CUDA 12.x for older GPU
+    # support (Maxwell/Pascal: 5.0, 5.2, 6.0, 6.1). This would allow shipping two versions
+    # of the build artifacts with varying backwards compatibility.
     # fermi     = "2.0"
     # kepler    = "3.0;3.5;3.7"
     # maxwell   = "5.0;5.2"
@@ -33,8 +36,9 @@ def main():
     # lovelace  = "8.9"
     # hopper    = "9.0"
     # blackwell =  "10.0;12.0"
-    cuda_arch_bin = "5.0;5.2;6.0;6.1;7.0;7.5,8.0"
-    cuda_arch_ptx = "8.0"
+    # CUDA 13.x dropped support for compute capability < 7.5 (Turing)
+    cuda_arch_bin = "7.5;8.0;8.6;8.9;9.0"
+    cuda_arch_ptx = "9.0"
 
     # NOTE: since 2.3.0 numpy upgraded from manylinux2014 to manylinux_2_28
     # see https://numpy.org/doc/stable/release/2.3.0-notes.html#numpy-2-3-0-release-notes