Contributes to #420 (hopefully "fixes", but we'll see 😅)

As described there, we observed wheel builds using scikit-build-core falling back to the GCC 8.5 at /usr/bin/g++.

  -- The CXX compiler identification is GNU 8.5.0
  -- The CUDA compiler identification is NVIDIA 12.9.86 with host compiler GNU 14.2.1
  -- Detecting CXX compiler ABI info
  -- Detecting CXX compiler ABI info - done
  -- Check for working CXX compiler: /usr/bin/g++ - skipped

Despite enabling the GCC 14 toolset:

It looks like the root cause is something like:

• sysconfig.get_config_var("CXX") returns the compiler the Python interpreter was built with
• at some point recently (possibly gh-148294: Make configure find g++ correctly python/cpython#148298), CPython started recording an absolute path like /usr/bin/g++
• we build CPython from source in the ci-wheel images used in RAPIDS, /usr/bin/g++ is the first C++ compiler on PATH at the time that build happens
• scikit-build-core reads that value when configuring CMake projects (scikit-build/scikit-build-core - src/scikit_build_Core/builder/generator.py)
• activating the GCC toolset doesn't work because that absolute path circumvents any PATH lookup at CMake configure time... /usr/bin/g++ is used unconditionally

This attempts to fix it by activating the GCC toolset before building CPython.

Notes for Reviewers

I haven't tested this locally, getting the PR up to start CI before I have to step away for a bit. Will try to return to it tonight.

I also think I have an idea of which CPython change caused this, will report an issue upstream tonight.