cuda.core: add C++17 standard flag and use -O2 optimization

Pass -std=c++17 (Linux) / /std:c++17 (Windows) for all builds,
and use -O2 instead of -O3 for non-debug Linux builds.

Partially addresses #1882.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: leofang

cuda_core/build_hooks.py

@@ -167,15 +167,17 @@ def get_sources(mod_name):
     extra_link_args = []
     extra_cythonize_kwargs = {}
     if sys.platform == "win32":
+        extra_compile_args += ["/std:c++17"]
         if debug:
             raise RuntimeError("Debuggable builds are not supported on Windows.")
     else:
+        extra_compile_args += ["-std=c++17"]
         if debug:
             extra_cythonize_kwargs["gdb_debug"] = True
             extra_compile_args += ["-g", "-O0"]
             extra_compile_args += ["-D _GLIBCXX_ASSERTIONS"]
         else:
-            extra_compile_args += ["-O3"]
+            extra_compile_args += ["-O2"]
             extra_link_args += ["-Wl,--strip-all"]
     if COMPILE_FOR_COVERAGE:
         # CYTHON_TRACE_NOGIL indicates to trace nogil functions.  It is not