Reapply "Plumb prebuilt_stages input through multi_arch_ci workflows" (#3934)

[ScottTodd]

Motivation

This reverts #3934 to re-apply #3856.

Technical Details

There were if: condition conflicts with 66b37d5.

I decided to also move build_pytorch_wheels_per_family after test_python_packages_per_family to match the non-multi-arch CI workflows (to keep rocm jobs together, then put framework jobs at the end)

Test Plan

• CI + pre-commit here.
• Multi-arch CI sanity check run: https://github.com/ROCm/TheRock/actions/runs/23013283219

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.